Slashdot Mirror
NT faster than Linux in tests
Mike_Miller writes "The lastest Mindcraft Study claims that Microsoft Windows NT Server is 2.5 times faster than Linux as a file server and 3.7 times faster as a web server. Their white paper shows that NT beats Linux on every test. " Anyone have a critique?
I notice it takes 10000 requests per child.. In the past, a lower figure was recommended in case of any leaks. I wonder if that could be a factor.
Check out the beginnings of a rebuttal on Linux Weekly News.
Well, i was under the impression that Linux still needs work in the SMP department... especially with more than 2 processors. NOT THAT I LIKE NT IN ANY WAY!! While I am suspicious of MS sponsoring the test, maby it is proof that Linux needs more work in the high end SMP department.
Coward
Linux 2.2.x has the same default window size. The memory limit is hard coded in Linux, unless you apply some patches.
i jsut looked at the graph the fact that they hit the linux box less prove the test wernt fair. a car going 2 miles an hour will always get there after a kid ridign a tricycle at 3 mph
Why no mention of where the bottleneck was on the Linux system? Was all processor time being consumed? Was the kernel snatching excessive cycles? Was the IO channel overloaded? The network saturated?
What exactly was the bottleneck, Mindcraft?
Maybe I'm completely off-base here, but weren't the network cards (Intel EtherExpress) they used the same ones which had poor Linux support because one company that made one of the chips on the card refused to release the specs?
Looking at what they listed as processes running before the benchmarks, it looks like they have apache being started by inetd. With the added overhead of having to start a new copy of apache for each incoming request, you would expect to see a huge performance drop. I also find it hard to believe that they could have accidently missed this while configuring the Linux box.
Hi.
I'm mcse certified to optimise iis. I couldn't dream of administering a linux system, even if I had the ability to turn a p-pp-pp-pp-age of a book by dr. seuss. You see- my arms were inadvertantly removed by dr. kervorkian when I was a witness to a mercy killing gone awry. I am also blind. My hearing is declining as well.
Would somebody help me please? I want to stomp out any need to ever again have to use telnet. It hurts my nose!
This post was written just as a flame bait/web site advert. He gives no solid examples to his biased openion. Anyone can make speculations like this. Please send this off to -1.
Remember that post a while ago that announced how 2600 posted some "proposed" hacks of what they would do to certain company's sites? Well, on the "hacked" Microsoft site they have a little headline that says:
Windows NT Server 4.0 Outperforms Solaris
Mindcraft, a Microsoft-certified testing lab, recently released a report that shows Windows NT Server 4.0 on a dual Pentium II/450 MHz system with 2 GB of RAM is more than 25 percent faster and offers 2.7 times better price-performance than Solaris x86 on a 486DX2.
Sounds familiar. Idiots...
"hi, i sell linux computers, and i think linux is good"
"hi, im payed by microsoft and i think NT is good."
"hi, i have a brain and i think you are both full of shit"
> and their latest work (e.g., SQL Server 7.0,
> Office 2000, NT 5.0/Win2000, IIS 5.0) has been
> of exceptional quality
What latest products? Where can I buy Office 2000 or NT5? or Win2000?
Sure, their vaporware has been of exceptional quality
(I see, I'm falling for flame bait)...
I agree. :)
http://slashdot.org/c omments.pl?sid=99/03/23/1318246&cid=1495
.. I agree this post doestn deserve +3 .. it's obvious flame bait..
proof?
Send this to -1
Either they're lying or they're telling the truth. If they're lying, replicate the tests and let Red Hat sue the testing company. If they're telling the truth, replicate the tests and figure out which part of the Linux code is too slow. Optimize it, release it. Either way, problem solved.
No wonder it was so slow. If I was going to run a high traffic site I think I would always keep 100 or so spare servers around.
Hey by calling my fellown AC blind, you have proved his point man.
Seriously, How does it feel like to suck bill gates? Does it taste good? Is this for kysh?
The net posts asking for help that are mentioned in the white paper appear to have been most likely made under the pseudonym:
will@whistlingfish.net
Use DejaNews.
I've been using one for several years, as have other people I know. The driver seems to work great...
- RF (dfelker@cnu.edu)
You mean kysh from varesearch on irc.debain.org #debian?
:)
He's a cool dude.. um.. person
shouldnt users be allowed to moderate the articles they read.. Maybe have that and the starndard modertaion figures.. So if anyone wants to see user based moderation they can just click that. As far as I can see this posting would go to the dump if it was user moderated.
NT's SMP is MUCH worse.. All the way from two cpus up..
If you read the test it says that there wasn't a problem with Linux, only with Apache.
Thanks for moving this down the rank of visually pleasing posts. I'm now proud of slashdot moderators.
Well, I don't know much about Linux, but it looks like they chose to use 4 processors because its SMP support is weak at the moment.
Does Apache support SMP out of the box, or do you have to recompile? Seems all they did was upgrade the kernel, not that you have to scour the web to find new versions of it.
Come one people. Most of responses to this comparison posted here are whiny and specious. The handful of responses that seem to bring up any real point claim that Linux's support for SMP and RAID is underdeveloped or immature.
Wouldn't you agree that a high end internet or file server should (1) have multiple processors, the more the better, and (2) a RAID array of disks to maximize file I/O?? Isn't this somewhere Linux needs to absolutely shine in order to be considered a server OS?
If this is war then Linux's defenses are strained. 90% of the business world that cares about this sort of thing is never going to notice or care about the "sponsored by M$" fine print or the slightly botched OS settings on Linux.
AC
...off to read the Linux press rebuttals (is there hope yet?)
Just a quick look at the details showed that they used almost every trick available for the NT config, and did almost nothing to the Linux config. They rushed this out to give NT admins something to hang onto -- if they couldn't twist ZD's arm into not releasing their results -- they just better have some footing when this gets crap gets blasted.
Novell was a victum of the Mindcraft/Microsoft partnership a few months ago. Mindcraft did a Netware v5.0 vs Windows NT v4.0 comparison.
f tcheck.html
Here is Novell's response:
http://www.novell.com/advantage/nw5/nw5-mindcra
Everyone can safely IGNORE anything from Mindcraft.
Chad A. Musson
Master Certified Novell Engineer
Microsoft Certified Systems Engineer
And I recommend Linux over half the time
They appear to have been made by:
will@whistlingfish.net
Which version of Samba were they running? I see that they "certify" that the tests were done with 2.0.1 at the end of the document, but the rest of the document seems to indicate that they were using 2.0.3. Is there anything wrong with 2.0.1 that has been fixed in the later release?
I didn't see the network card!!
What NIC are they using? They also dont mention the network config.. I assume it was gig from the server to switched 100 to the clients.
I've seen Linux hot cache at 490Mbit/s across a netgear gig card.. (But that was a 21264 with 512Megs of ram)..
It seems that they dedicated a processor to each card under NT, but they failed to mention if they even enabled the other interfaces under linux.
err... So let me get this straight. These fsckheads who had no idea what the hell they were doing with Linux, tried to get help from RedHat, probably went on newsgroups and asked dumb MCSE-type questions and didn't get any answers, yet they claim they "optimally configured" both NT _AND_ Linux... hmm. I smell horsesh*t.
Go to the main page:
http://www.mindcraft.com/
And see how crappy Solaris is and NetWare supposedly are in comparison to the Great Windows NT. You cannot believe everything you read, and if you think you can, read a David Hewson column.
Keep up the good work mr Gates. I am glad that you lead and keep innovating the industry and the os market. I would like another test showing how well smp is handled compared to linux. Thank you Mr Gates. I can't wait to fire these idiots who put solaris and linux on our corporate servers. THis article also shows that the oss movement is really just a fad like java and the network computer. Its too bad that Linus can't program for squat. Its great that at least Bill Gates can write code. Just look at what he did with ms dos.
Okay.
So let's take their little report as gospel and not worry about the details.
Someone *please* explain to me why it is that sites like Dejanews (Linux+Apache) and Yahoo (FreeBSD+Apache) are always up, and always fast, yet sites like microsoft.com and eBay (which use NT and IIS) are constantly coming up with server errors, screwed up pages, and are slower than molasses running uphill in the winter?
Did anyone else notice that they also have "NT 10 times faster than Solaris" and "NT 25% faster than Netware" articles up there? Gimmie a break.
Explain to me why it is that they converted hotmail.com back to Solaris after NT failed to hold up under the load if it's that much better?
All I worry about with this is that it will give fearful MS-loving managers something to wave around. You gotta admit it's written very professionally, and appears to be quite thorough unless you are a tech person who is familiar with both OSes and all the software they are running.
If the Linux box didn't get the full gig, then why didn't they just pull a couple from the NT box?
I remember reading about biased Netware vs NT tests being recalled in January, done by Mindcraft. IIRC, they had similarly outrageous results of NT blowing Netware away, and those results went against anyone else's benchmarks on the subject.
"Microsoft's goal is to make money. In order to attain this goal, they must have a superior product in order to get people to buy it."
f tcheck.html
This is not necessarily true. Their goal is to make money, and to attain this goal they spend billions on marketing an image that unquestioning buyers accept. They sell most of their products preinstalled, and acheived the market share through exclusive distribution channels (i.e. OEM strangleholds) that prevent the entry of competition.
"...their latest work (e.g., SQL Server 7.0, Office 2000, NT 5.0/Win2000, IIS 5.0) has been of exceptional quality..."
Three of these four products have not shipped. This reminds me of the marketing I mentioned earlier. However, I realize that Microsoft is trying very hard to get better. Unfortunately, the way they are going about it is all wrong.
Paying Mindcraft to do Linux/NT comparisons when they don't know the first thing about how to properly compile, configure and tune the Linux kernel, or how to configure Apache properly, or how to ask for or find any help on these things, is really sad. Mindcraft has done this before with Netware, and will probably do it again the next time Microsoft gets threatened by something clearly better and faster.
http://www.novell.com/advantage/nw5/nw5-mindcra
Good point... did they use maximum optimization when they compiled the kernel, apache, and Samba?
Apache 1.3.4 is what the white paper said they were using.
But the newsgroup post says they were using
Apache 2.0.3... (As some one said earlier,
the user agent for newsgroup post was ie5 on
NT 5.0) I wonder if they are just using
software from the future? It's hard to get
newsgroup support for that sort of stuff.
Those peak values... by iabervon on Tuesday April 13, @09:41PM EDT (User Info) http:// According to Apache, they stopped tuning for performance when a low-end machine could saturate a T1. The Linux box in the test peaked at 27Mbit/second. That is, the Linux box could saturate 10base-T, let alone anything reasonably between a web server and the clients. If you've got multiple T1s coming in, you may as well have multiple machines at the end. For file service, the Linux box put out 114Mbit/second. That's more than 100base-T, and a significant chunk of a gigabit. If you've got this sort of thing on your network, file server performance is the least of your worries. Unless you intend to have a separate network for file service, like in their tests, this isn't going to be a big deal. I've got that sort of thing on my network. Sun E3000's running Solaris, with 8 100MB interfaces.
Their server is slow as hell, i can't download the whole article. Seems that they need to upgrate to linux and apache ;)
You can do it. It just depends on how much you want to spend. We've done it multiple times.
This is obviousy pure FUD.....
Every study shown on thier page is pro MS
read this story at infoworld;
http://www.infowor ld.com/cgi-bin/displayStory.pl?990413.ecnt.htm
Looks like MS past the denial stage and finally telling the truth.
Choice excerpts:
Such measures are needed because Unix-based platforms have proven to be much more robust than NT.
"Unix systems are generally a lot more solid and don't tend to go down as often," Thomas said. "Some haven't been cycled for years."
The two biggest complaints about Windows NT -- the need to constantly reboot servers and systems failures that lead to the blue screen of death
Microsoft hopes to restore some of NT's tarnished reputation in the enterprise
"I'm astonished that people run their business on a system [Windows NT] that may not be awake for the week," said Anne Thomas, analyst at the Patricia Seybold Group in Boston. "It can take up to 30 minutes to reboot the servers, databases, and applications."
If Mindcraft is independent, maybe one Linux distributor should sponsor them to do the same test again. This time let the distro's engineers with help from Linux community do the tuning. If Linux in the second test performed significantly better than the first test, Mindcraft and MS will be heavily discredited.
Hey M$, you look really scared...
C'mon M$, hit us, hit us with all your strenght, make us mad, make us really mad, make people like us, make people hate you, make us stronger and then...DIE!
The reason any truely seasoned IT professional
scoffs at this document is because of the source.
Microsoft has a long past of lies and
misinformation used to its own ends. The idea that
Linux can't do SMP and Raid at the same time is
completely illogical. I don't know that it can for
sure mind you, but experience in the industy
for over a decade leads me to believe that the
statements made by Microsoft are untrue.
I use DOS, Windows 9x, NT, BSD, and Linux in the
field and in my experience DOS lost alot of steem
with the advent of 95. NT has proven to be quite
sluggish as well due to patch after patch to
repair poorly written software that was rushed out
the door at the expense of the user.
To further degrade Microsoft's findings, if microsoft didn't lie again, how many webservers
run on 4 or more processors with Raids? If you where to pay that much for a computer why not buy
a high end unix server that creams NT? Why run a
GUI 24/7 on a computer, that if operating perfectly, should not be seen for a year or so?
NT was allready tested independently against many
operating systems, and it *was* the bottom rung.
I don't believe Microsoft, nor any paid
"independent" lab.
They also seem to have not changed the size of the tcp queue, as they did with NT, and for some reason compiled things with 486 optimization, rather than 686 optimization. They also limited apache to "MaxSpareServers 290," "StartServers 10," and "MaxClients 290" rather than, say,
MaxSpareServers 1000
StartServers 100
MaxClients 5000
Wankers
One thing I will say is that finding good documentation on 2.2.x is a big pain. Documentation in general is a big problem.. All of the HOWTO's on Sunsite are quite out of date. I had to search for quite a while to find good information on GRE tunneling in 2.2.x..
Come on, Slashdotites. It's one thing to rip
on the survey for valid reasons, but this knee-jerk speculation here is doing nobody any good.
They used those crappy ZD benchmarks.
Actually, IIS on NT handles more big sites than Linux does. MSNBC is the number one news site on the net, whomps slashdot by several orders of magnitude, and it runs on IIS.
Well, linux documentation has always beena bit shitty, no supprise there, since when can programmers/hackers with failed English levels be bothered to write a decent html help file with pics/tables, and a 200k readme doesnt cut it.
Seriously, if you dont like the results, do your own tests on two high powered machines, but dont cry if the linux box is slow.
Mindcraft page didn't say they expanded the TCP port range (/proc/sys/net/ipv4/ip_local_port_range). I'm sure Linux would ran out of TCP ports during the Web benchmark.
good post
Aside from it being blatant FUD and pure BS, this little disclaimer just about sums everything up in itself:
NOTICE: The information in this publication is subject to change without notice.
MINDCRAFT, INC. SHALL NOT BE LIABLE FOR ERRORS OR OMISSIONS CONTAINED HEREIN,NOR
FOR INCIDENTAL OR CONSEQUENTIAL DAMAGES RESULTING FROM THE FURNISHING,
PERFORMANCE, USE OF THIS MATERIAL.
Here are a few things that were mentioned earlier but seem to have been overlooked:
:)
/sbin/kerneld, /sbin/mingetty [on tty2, tty3,
:)
/.-proof account at slashdot-me@altavista.net
* From Andrew Trigell (original Author of Samba):
>They set "widelinks = no" now I wonder why they did that
>In case you haven't guessed, that will lower the performance enormously. It adds 3 chdir()
>calls and 3 getwd() calls to every filename lookup. That will especially hurt on a SMP
>system.
* Samba and apache were run from inetd! I'm no apache guru--but wouldn't this prevent the preforking thing from happening? Lack of a thread pool should kill performance.
>>The following processes were running immediately before the NetBench and
WebBench tests: init, (kflushd),(kpiod), (kswapd),
syslogd, klogd, crond, inetd, bash,
tty4, tty5, and tty6], update (bdflush), and portmap
Where are smbd and httpd ???
Moderators: previous posts that mentioned these issues weren't noticed. Notice me
Send email to my spam-proof,
So does this mean I can put up a web site and blatantly lie about anyone or any business I don't like (maybe spread some nasty lies about a competitor's business?) and put that disclaimer on the end and not have to worry about getting sued?
What are you kidding me? And this from an AC?
Probably not stupid, just configuring it that way on purpose... after all they said they couldn't get *any* help from any newsgroup or red hat. I am sure careful thought went into configuring this "test" to get the figures they wanted (assuming of course they didn't just make the numbers up).
Yep and Win95 is faster than Win 3.11
NT 4.0 is faster than Win 95
and Win 98 is faster than NT 4.0
And Win 2000 will be the fastest of all!
Some people will buy anything, so long as there is the Micro$haft seal of approval on it!!
If you actually take time to read the article, you'll notice they limited the NT Box to 1 GB.
Ah yes, and slashdot is running on such an incredibly huge server too. I don't think you can compare whatever Server msnbc is undoubtedly running computer to the little PC that slashdot is running on. Not to mention that we don't know how many hits MSNBC gets compared to /. (maybe you would enlighten us as to how you know the servers and # of hits each site gets?).
We all know that this is crap. Are all sysadmins running solaris or linux/FreeBSD stupid? I think not. Only die hard NT guys will agree that the Mindcraft tests were done correctly.
I hate how Weiner whines (pun intended) about searching the web for the latest Linux drivers and how NT has them all bundled. Give me a break, I spent quite a few hours looking for NT server hotfixes, patches, correct and working drivers. Most admins who use FreeBSD or GNU/Linux know where to get the latest stuff anyways.
I never used ZDBOp or the other tools but who knows if they report fair marks. Reminds me of ZD not supporting AMD/3D Now and still comparing to Intel's chips.
I wouldn't doubt that they pulled the now famous switcheroo and put in a faster machine. Next thing you know they will say the current USB is faster than Firewire. OK. I'm gulible, I will take 10 copies of NT server please.
FreeBSD or GNU/Linux all the way brother.
We might as well get used to this because it's only going to increase as we closer to the mythical ship date for Not-There 2000.
The fact of the matter is that M$ sees Linux as an entrenched enemy that first needs to be "softened up" with artillery through friendly press and anal-ylsts. These are the same people (NT bandwagoners) who have seen their cushy gigs trashed by the popularity of Linux. Some see the light and hop on the bandwagon while others seek to strike out.
The DH Brown report coupled with MindCruft and the coming bevy of "main stream" trade publications performing "benchmarks" are clearly attempting to discredit Linux as an "enterprise" OS. This is why all the latest benchmarks are being performed on large systems (usually 4-way SMP) with a couple gigs of RAM and a large RAID arrays. It is also why DH Brown chose to compare $49 off-the-shelf operating systems with $$$ enterprise server configurations
They know they'll get their butts kicked on the low-end and are trying to build a beach head at the high end where they have strength, but also where the customers are pissed about having to reboot their systems continually.
Get used to it. The war has just begun...
That one is easy to explain.
If you install IIS 4.0 on Win 9x or NT 4.0 WKS it becomes Personal Web Server with an 10 user limit.
Please substantiate.
This is the same Mindcraft that said that Windows 95 was much easier to use than a Mac. They are pretty much M$es PR bunch.
I think you should do your own tests whether you like the results or not. Let the best OS win.
What an idiot! Here we have a guy who in the same post is claiming Microsoft is trying to cause FUD while simultaneously repeating that tired lie about MS supposedly trying to move Hotmail to NT and NT not being able to handle it. Amazing. There is far more anti-MS FUD out there than MS spread FUD. Open your eyes!
It would be fun to have a small group of gifted tuners set up a pre-agreed upon hardware configuration for a match against identical hardware running NT. Perhaps someone will challenge the MindCraft results and offer to do this? Have it agreed upon that the results would be released in a followup whitepaper.
BobP
Perhaps redhat should get of their ass and configure the default installed system to be a little more optimized WELL!!, not for 486's
Not to mention also how slow RH is to release a2.2 system yawn! the wheel turns slowly
Interesting then why MS had to abandon moving HotMail to NT from Solaris because it couldn't handle the load!
So Netscape dies trying to run a little Java applet and you blame Microsoft's web server? Sorry kiddo, your browser is busted. Run a real one.
I don't know. That's questionable. What I do know is that my last post was entirely unfairly moderated down for the fact that I criticized the moderators as well as the poster of that post for his viewpoint (and theirs in moderating him up). If one can't even express such sentiments civilly on /., then I guess I see why so many people post as AC, and get pissed about the way /. is run.
Fnkmaster (posting as AC so I don't get perm moderated down to 0, and considering bitching to Rob about his moderation system)
If they had a clue, they would use libc6 threads, not forks, forks suck ass, any programmer knows that real threads are much better tohandle shit.
All this config for X spare server, Y min servers is bullcrap!, why cant apache be smart andknow how much it needs it self depending on the amount of hits its getting. Get of your ass coders, and code something smart/dynamic/self learning.
Get out of the 80's mentality of configuring 10000 options to get something to work, make the code SMART, not DUMB.
Somebody dropped a huge wad of cash into their lap. This is not unusual business practice, it is called bribery.
Get a decent ISP you god damn looser, its fast and isntant for me. WANKER
"Mindcraft, Inc. conducted the performance tests described in this report between March 10 and March 13, 1999. Microsoft Corporation sponsored the testing reported herein."
Hmm, that looks like a good optimization.
If you read the test it says that there wasn't a problem with Linux, only with Apache
Problem with Apache my ass! Apache is the most used/tested/hacked/twisted http server around. It beats IIS or any other crappy pseudo-server-security-hole-inside you can find on the market. Apache is number one.
You are wrong. Linux requires patches to use more than 960MB of memory.
Sure, I'll put 1Terabyte of RAM in my machine and just pass LILO a mem=1000...00M's lines...ya, right!
Of course there's an upper limit you idiot.
Hmm... one area where 98/NT scores better than Linux is net config on initial install. Nothing is worse than having a computer with a newly installed OS and no net driver for your particular card. Cards come with 98/NT drivers on a disk -- problem solved. For Linux you must connect and go to a website to get the proper driver if doesn't come on the distro CD. I guess that's one reason companies like Red Hat make money off of this whole linux thing anyway...
AC
I could be way off here, but don't pentium II's only cache 512mb maximum? That might explain the performance hit, if the beowulf consisted of pentium II's that is. Maybe I'm thinking about the older klamath pII's...
[asimov] [/home/navindra] uname -m /proc/sys/fs/file-max
i586
[asimov] [/home/navindra] cat
4096
[asimov] [/home/navindra] limit openfiles
openfiles 1024
Certified
Shills
Extraordinaire
"A ferrari doesn't run well in a foot of mud. Let's stick a tank and a ferrari in a foot of mud and see who's faster!"
Not that NT is a tank. Tanks don't blue screen.
hmmm is it openfiles or descriptors that matter here? *shrug* either way both seem to default to 1024 under tcsh.
really mean, but really fucking funny, that.
From a highly trafficked SMP Linux web server...(80+ million transactions per month)
MaxKeepAliveRequests 100
KeepAliveTimeout 15
MinSpareServers 20
MaxSpareServers 30
StartServers 150
MaxClients 500
MaxRequestsPerChild 3000
uptime...
1:38am up 27 days, 2:43, 1 user, load average: 0.00, 0.02, 0.05
Last I saw it was running qmail on solaris,
let me check this.
Received: (qmail 57854 invoked by uid 0); 14 Apr 1999 05:40:30 -0000
Well its running qmail, I did see solaris, but I cannot remember where.
"pointy-haired"?
If Microsoft really wants to show off their product VS. Linux, the tests should be conducted "50's HotRod" style. Let Micrsoft choose a PC Vendor -or- to perp the NT Server, then let the Linux Community select a group to oversee prepping the Linux box. This way we know that the Linux box will perform to its potential.
If both parties are willing to publish their setup specs (so Microsoft doesn't slip in some "non-production" kernel or whatever...), there is no question about the outcome of the tests. Let a mutual test lab conduct the tests and see what happens.
...but Microsoft won't let that happen! They no that there is no way that the GUI-infested kernel of NT can hang with the raw power of a *NIX system. Hell, I'd put HP-UX, or even IRIX up against NT and NT would never have a chance in hell. NT is a nicer because of the instant GUI setup, but only SYSADMINS that got certified by "Sally Struthers school of Home System Administration" would benefit from it. I've used NT for many years, cried when it unexplicably crashed, and gnashed my teeth when it chokes on large print jobs.
You see, any company that uses a *NIX, has a sysadmin who probably knows a good deal about the operating system. This is necessary because a 16 year-old GUIphyte could install and setup NT, but would be lost as a Baptist preacher in a whorehouse when the system crashes. There is some thing very good about an OS have a steep learning curve. It may take longer to learn and use, but your knowledge is much more intimate, than "Click here", "check this box", and "reinstall NT from the CD...".
Eric
Sorry, Win 98 absolutely stinks when it comes to using it's own drivers w/ network cards. Case in point... Intel EE card, every other boot Win98 would complain about not having the updated drivers, when win98 is newer than the EE card. I finally yanked the card it was ticking me off so much. About once a month win98 has the same problem w/ the 3c509 card.
That's funny though. My RedHat 5.2 runs Quake, telnet, FTP server, Router to my Games boxes(windows, ick) and Apache and I dont have any problems with over 2000 hits per day.
Oh, BTW, did I mention this all runs on a AMD K6 166, with 32 megs?
Try that with windows 98 buddy.
-=A.C.
This benchmark is very important for a whole lot of reasons. Several are obvious (PR milage, FUD, etc), but several more are very difficult to analyze and determine the merits of.
First, to the responses. I was very dissapointed to see that so many dismissed this or were downright hostile. This is data. If the best you can do to refute data is to yell and throw things at it, you should consider backing some other horse, because this one don't want your help.
On the purely technical side. I have a friend working with this very platform, so I happen to know that that specific Linux MegaRAID controler driver is shit. It comes with a stack of warnings about how it has "not been tested for performance", and is not Open Source.
On the business side, think of the genius of this benchmark. Microsoft and its supporters wish to cast the up-and-comer as a poor performer, but Open Source is too powerful a development model. They can't out-develop us, so instead, they perform a test which is really demonstrating that one closed source application is slower than another closed source application, but... and here it comes... one of them was caught red-handed running under Linux!
Now, how do you debate this. You know that for the next six months you're going to hear friends, business contacts, etc. saying, "So, I hear Linux is half-fast compared to Windows"; "Maybe we should go with a proven performer"; "I just can't afford the performance hit to run Linux".
This is the problem that must be solved. Part of the solution is looking at what the box that was described in that benchmark would cost. Now try running that benchmark again... on a $1000 system! So, how many $1000 systems can I buy for the cost of one of those Dell beasts that ship with beta drivers? Got it.
On the other hand, you could take the monetary and more statistical approach. The graphs on that site show a sharp plataeu that other Linux benchmarkers have not seen. Why? What single factor (and with that sharp a leveling-off, it's likely to be a single factor) caused the Linux box to choke?
Then again, there's the empirical test. Does the installed base of Linux-based servers blow the doors of off Windows servers at the same price point? Why, hell yes! Some reports put the *BSDs higher, but certainly both BSD and Linux blow the doors off of Windows in the average installed environment, so Mindspring must have done SOMETHING wrong (or right, depending on their goal).
Food for thought.
-AJS
If Mindcraft wishes to salvage whatever remaining credibility they have, someone in management there will see to an immediate retraction.
The baldfaced bogusness of this test is what is most laughable. They could have at least faked a test with only slightly higher scores for NT over Linux.
Send Mindcraft management email and tell them their reputation just hit the shitter.
See http://linuxtoday.com/stories/4938.html
Yes, and if you have read I think its the Halloween Documents, these graphs look remarkably like those Microsoft have already paraded!
> You can't use more than 100 threads on IIS 4.
Wasn't the study was looking at the number of CLIENT threads not server threads? How would it know how many threads IIS was using anyway?
There are some comments on this, see http://lwn.net/1999/features/MindCraft.phtml
If you want to know the truth - it looks something like this: (not that anyone really cares)
Win 3.x is a DOG if you try to run 32 bit code (any semi modern app) - both 95 and NT are quite a bit faster in this respect. If you want to run 16 bit code, Win 3.x wins.
Win 95 is SLIGHTLY faster than NT4 but NT4 is much more stable - unfortunately, if you run NT, forget about running most of your favorite games. Also, NT generally does not work well on notebooks computers.
NT3.5x is faster and more stable than NT4 but has the ugly Win 3.x GUI.
Win 98 is an absolute DOG compared to either 95 or NT and is MUCH less stable in both cases - and the Win 98 GUI "enhancements" are especially horrid.
NT5 (Win 2k) will be slower and less reliable than 95 or NT 4 - it will probably be more reliable than 98 but with the same horrid GUI "enhancements."
Practically any UNIX system - even microkernel based UNIX - will outperform any of the above listed Microsoft products on equal hardware. You really need not look any further than the Microsoft owned (but not created) Hotmail system where they tried and failed to replace the FreeBSD/Apache web server frontend and the Solaris backend with NT/IIS/etc.
Actually we should raise money for this, get two
identical systems from Dell or Compaq.
Then challenge MS to send their best engineers to
a public face-off.
The results to be audited by one of the top
accounting firms like Ernst and Young etc.
The important factor - to prevent fudging
a copy of the OS will be bought at a store
and will be installed on a vanilla machine
in front of auditors. then it will be tuned
in front of auditors - then the tests will be run
again in front of auditors and the public.
MS should be challenged to this. Publicly.
This needs to be settled right away
so that they will not try such outrageous
and obvious manipulation of the truth.
I would like to start organizing a drive to
do this - need financial contributions, contributions in kind
( machines/software auditors time and organizational contributions).
If you'd like to help
send e-mail to faceoff@borwankar.com
Nitin
----------------
Debian is much better than RedHat
If a sysadmin shouldn't be expected to have to apply patches, then they shouldn't be expected to have to edit the NT registry either. It looks like this company knows quite a bit about NT but not much about Linux at all.
1024MB != 960MB.
Hmm. Two damaging reports on Linux in as many weeks. I don't believe in coincidence, so Microsoft must be gearing up the war machine to take on Linux. Look who sponsored this survey. On the plus side this means they're taking Linux very seriously and believe it to be a threat. On the minus side, this means that they're going to be playing all their usual dirty tricks. Quietly sponsoring a survey here, an article there. You think Bill told someone "You know, a survey showing that NT is twice as fast as Linux is just what the doctor ordered right now..."? It's been known to happen. It's pretty easy to misconfigure Linux to run like a dog. Same goes for NT. Does anyone out there have the resources to implement these tests with a correctly set-up Linux box?
My solution to this - and you would be wise to do the same regardless of weather you use Linux, WinXYZ, or anything other OS - is to research your hardware purchase very carefully. Sure, any old bottom of the barrel NIC card comes with Win 95 drivers on the disk but very often they don't work well at all and good luck finding driver updates! For NIC cards in particular, I tend to buy 10/100 PCI cards based on the DEC 21140 (specifically I like the Kingston KNE100TX) - why? Well, it's about a $45 card and it works on practically anything - Windows 95/98/NT, FreeBSD, Linux - without even installing a driver! (That's right, the OS comes with the driver built in.) Hell - that card will even works in a PCI PowerMac! OTOH - if you're a fool - you will buy the $30 Kingston KNE110TX which is based on a proprietary chip and doesn't work on anywhere near the amount of systems - moral of the story - save $15, gain extended pain in the ass.
When ZD used them, they found that Linux beat NT, and IIRC this was with stock RH5.1, OpenLinux 1.3 and SuSe whatever-the-latest-version-was. So don't blame ZD for Mindcrafts bias.
Uh huh - that's why MSNBC (and any other "big" IIS site) runs on a cluster of many extremely high end NT boxes and slashdot runs on a single Linux PC!
That's why ftp.cdrom.com can churn out terabytes of data per month and maintain 3500+ concurrent connections from a single PC (with a large RAID) running FreeBSD where Microsoft uses a cluster of extremely high end NT boxes to operate their own FTP site.
That's why the Microsoft owned Hotmail system runs FreeBSD/Apache web servers and a Solaris backed. Not NT!
And the clustering has just as much to do with server availability as it does handling the connect load. I know this first hand - for some time I worked for a service provider that ran NT systems. They started with one dual Pentium NT machine - the load was fairly heavy but that one machine could keep up. The problem is, that one machine would blue screen several times a week, which instantly resulted in the admin receiving angry phone calls from important clients. Their solution was to but more NT boxes and build a cluster - not because they needed to handle additional concurrent connections but because they were faced with the prospect of loosing clients if they couldn't make the system more reliable.
After reading the 'white' paper I conclude that: .inis etc
Alot more tunes were done to the NT and IIS eg reg keys and
than were done for the Linux setup
and I just dont believe those results, the numbers are just too big!
Maybe RH can sponsor some 'inder-pendant' tests also as M$ does for propper-gander
The proof and truth are in the real world
how many more people actually use NT for FS or HTTP on the net
what about M$hotmail...- if is was good they would
QED
i can read any other site really fast.
Besides, you checked it out two hours more than two hours later...
They're missing a basic tenet of GNU/Linux: you use what works best; a webserver for serving static images would obviously be best with FreeBSD, not Linux.
flamebait. let's flip a 180 and say it again.
-d
ask linus to put his own quad Xeon to the test...
this is a legacy option from a long time back. there aren't any known memory leaks and it is quite safe to set this to infinity. this is the way our servers are set and several of them serve over a few dozen gigs a day...and keep doing that for several months straight.
no memory leaks noted.
so please keep current with documentation and whatnot.
-d
I just noticed the timing of the announcement of this "study".
:-)
It's dated April 13. The testing was done a month ago. Took that long to write up?
Well, it's April 14 now. Today, in Atlanta, at the AIIM '99 show, there will be a panel featuring Linus, Jon "maddog" Hall, Tony Iams (the DH Brown Linux report author), and Jim Ewel of Microsoft who I think is the marketing guy for SQL Server 7.0 (might be marketing guy in W2K now).
Also, Comdex starts soon afterwards on April 19. Bill Gates and Linus will be giving the keynotes. This Comdex will also be Win2000's coming out party.
M$'s PR coordination ain't too shabby
They comment in the report that they attempted to find good documentation and help regarding their setup. We have people here who have dug up their requests for assistance. Were they given any?
If we are going to convince businesses that this user community support thing works, then we need to figure out how to make it work. There are people out there running Very High Performance web sites. Do they publish their tips and tweaks? Or are we running into a situation where the people that have the access to these highend machines are shackled by corporate policy?
Seeing Intel's strong interest in Linux on Quad Zeon boxes (that they have demoed several times) it would seem there are some people that could start a High Performance Linux web site with poignant information about what to do and what no to do. I do not have the base line information for such a site (my biggest Linux box was a P-Pro w/ 256MB and 54GB).
I would be happy to coordinate with people who have hard data to get it up on the net and in discussable forums (prob. use the slash code to boot!).
Let's do it...
Chris
chris@pugrud.net
-- My slashdot login is here somewhere..?.
even beter, check out
s t=www.microsoft.com
http://www.netcraft.com/cgi-bin/Survey/whats?ho
It looks like Micro$oft wants us all to go to Win2K, but prefers to use NT 3.x itself...
Tell me..
Why does altavista NOT run NT?
what about netscape. which is said
to be the most heavily trafficated website
in the world.?? no nt there.
check out www.netcraft.com and you will
see which sites run what.
...in my house I have 10 machines, 6 of them are a prototype Beowulf cluster and the other 4 are some flavor of WinBlows. All of them are running the distributed.net rc5des clients. The Linux machines process more blocks per machine, even the slower 486's do 5 blocks every 24 hrs, while the windows 233 machine only do about 10, while the Linux 200 does 25, and the Linux 200 is my production machine, which means it is cranking blocks while I compile new kernels, play Quake and surf the web (sometimes all at once), Its load is almost always between 1.5 and 2. If I so much as want to check my email on a windows machine, I have to shut the rc5des client down or the system will crash. The Linux box's also go months without a reboot, while I have trouble keeping the win95/98/NT system's up for more than a few days. On top of that I never have had to reinstall Linux because of corrupted Operating System, while I have to reinstall windows on one box or another every few months. What all this says to me is, if I'm in a lab and I need to run NetBench 5.1, I'll choose NT, but if I need to get real work done, I'll choose Linux.
I would also like to point out that there was no mention in this study of uptimes/reliability or cost of ownership/performance. We all ready know how those tests would come out.
Read the kernel BUG log. There are some TCP problems in 2.2.2 - 2.2.3. So selecting them instead of 2.2.0 or 2.2.4 is quite intentional
So the results may be quite right overall... (for the experimental setup).
I don't know if it was the same product, but I answered a MICROS~1 survey once. The question were *so* loaded it was hilarious.
Questions like:
* Do you value the integration of an application with the interface and conventions of the operating system? (right answer: yes)
* Would you prefer a well-intergrated application, or one meant to work exactly the same on multiple operating systems that does not take advantage of the features of your operating system? (correct answer: like the well integrated app. Don't like java.)
* etc. It was crazy. I don't know why they even bothered.
the client side testing, ZD scripts are most probably OK, why should anyone try to cheat here.
the NT server stuff goes into detail on setting up all four interfaces -
the linux box saturates at the value expected for a single interface, and there are no details about setting up the other three - isn't only one interface active by default?
so the NT test should give four times the throughput, but doesn't
The very standard and obvious answer to cope with this is to chroot() the samba server (which you must do anyway if you absolutly don't want risking samba bugs exploits exporting the whole filesystems).
In any sound benchmark performance study, this should have been taken in account, and figures for both safe and "unsafe" (or chrooted) performance should have been given. Much as in SPEC CPU benchmarks (compiler with standard optimisations, compiler with any optimisation, specific flags).
You should probably mail directly to CmdrTaco, so that he releases logs/statistics, tells developpers what tools would be useful for him to monitor performances, or otherwise generally cooperate with Unix/Linux free software developpers. (if it is not already done).
The only reason for why apache would be the bottleneck is if it's REALLY REALLY malconfigured
www.altavista.com is running AV/1.0.1
Now, i checked this a few weeks ago, and i remember i was puzzled why altavista.com ran NT. Anyway..netcraft.com shows AV/1.0.1 (AltaVista ?)
altavista is Compaq-isht, and Compaq is official MS-partner, i even saw a commercial stating altavista.com running on NT.
i-don't-post-that-often-to-care-to-register
... and a very interesting thing about this graph is that a number of different servers (including apache, roxen, boa and others) have performance which drops dramatically at a level of about 125 simultaneous requests - similar to that reported.
(that's the graph mentioned in the previous post).
whether this is caused by the fd maximum of 1024 or not is left as an exercise to readers...
-duncanNear the end of the article, Mindcraft/M$ admits to not fully tuning Apache and Linux.
:(
But they also say that they looked and were unable to find concise, useful information.
Me thinks this survey is more a comment on the support offered by the Linux community than on the performance of Linux or Apache.
After all, I have a hard time believing that Apache would run slower than IIS4 under normal, tuned conditions. I should know, I crawl -er, run- a couple small IIS servers
(Autobiography: college student w/ two years NT and FreeBSD admin experience. Prior to that, no network experience and used DOS/Win3.1/Win95)
Can't we find someone that works on IIS for a living that's willing to setup a challenge with some 'experienced' apache tuners?
Seems that the stakes would be very high -- if NT were to win we couldn't say 'well it wasn't setup right, yada yada....'
And if Linux wins then MS will just lie about it anyway and tell us how the results to 'accurately represent their customers intended uses of iis.'
>and presumable still do
Yup, they do. Here's Novell's rebuttal to a similar 'independent comparison' of NT4 and Netware 5 http://www.novell.com/a dvantage/nw5/nw5-mindcraftcheck.html. The study that prompted this particular article was done by none other than mindcraft, so I wouldn't put too much faith in the results of any of their tests.
Kathryn.
Another ./ poster suggested somewhere further down
that they maybe 'forgot' to enable SMP support
when compiling their new 2.2 kernel... which
sounds very plausible.
Agreed!
I use slackware for my servers specifically an older kernel for stability and speed. My older slackware distribs blow away rh5.anything for server performance and adhering to the posix standard for filesystem layout. RH5 is great as a desktop distrib... as a server it really bites and needs to be re-designed so that a server install would run as fast as older slackware releases. and this crazyness of having one server with massive ram and hdd space is plain stupid.. use a cluster of servers for reliability and load balancing it's cheaper, easier, and makes the server closet look damn impressive!
"Why is our intranet down for all of the Southwest?" " because our dell 900processor 22googlemeg ram, 9054 hetagig hard drive server was unplugged by the cleaning lady"
Having one machine responsible for a gob of data and bandwidth is plain stupid.
Nobody seems to notice the optimalisation options given when compiling samba.
they used -O4 -m486 !
I believe that gcc-2.7.2.3 only can handle -O2 -m486 as a reliable optimalisation
...but maybe I'm wrong....
greenthings
roger
# queso altavista.com
204.123.2.97:80 * Berkeley: usually Digital Unix, OSF/1 V3.0, HP-UX 10.x
OK, so now we all know that the test was almost certainly rigged, what needs to happen is for an *independent* party to perform exactly the same test with advice from Apache or whoever on how to configure the webserver for a level - field head to head. The benchmarks also need to be certified by ZD.
Why let MicroFUD and MindBogglesCraft get away with it?
If you go to http://slashdot.org/users.pl?op=userinfo&nick=spec tecjr you will see that he has consistently posted messages in support of Micros~1 - along with a disclaimer that he doesn't speak for them.
Ummmm - so that's why /. is number 38 on the hot100 site? Cause it gets so few hits? Grow a brain and give some backup to your statements. How many times have I gone to one of the sites 'ruled' by IIS only to get an error message?
booyah baby, im so good
The reason everyone is pissed off is because:
A. Those who actually have to deal with NT as a web server know that it is not faster and is not as scalable; ever wonder why you get all those errors from web sites running IIS?
B. Idiots who say 'well go test it yourself' miss the point - the moronic suits and press will swallow this M$ crap whole - they won't test it or question who pays Bruce Weiner @ Mindcrap.
C. After the antitrust trial with the M$ 'tests' - how anyone can believe anything from a M$ sponsored test lab is beyond me.
i noticed two things about the report :
kernel:
The Linux kernel limited itself to use only 960 MB of RAM. Umm.. looks like they forgot to add the
append = "mem=4000" line in lilo, so while the NT
machine was using 4Gb, Linux was only using 960Mb.
Apache:
They have the min spare servers set to 1 but
the maxclients/max spare servers set to 290, this
is going to cause apache to spawn a lot of processes and keep them spawned, using up resources and lagging the system. For the type of
tests they were doing the probably should have had
maxclients set to whatever the max was for the memory based on the processes they were running,
set max clients to a few under that and min clients somewhere in the middle, and reduce the
amount of requests per client.
This test was definately a fraud, they probably
optimised the NT machine to the hilt, while
their lack of technical ability is apparent.
also note that the LOG commit to disk for IIS
was set at 100 MegaBytes. This means the IIS
probabely did not write to the disk until the
test was over. Apache by contrast, writes much
more frequently.
Now that I think of it, that must be the key,
since a 100 Megabyte write would take up a
substaintial portion of the test. I also recall
the WebBench test only lasting 300 seconds.
Is it just me or is there a problem with a server with 4 100mb cards getting 411mb throughput. I think they should look at local caching.
I tend to agree with you on win98 But I have to disagree with you on Windows 2000.
I recently loaded Beta candidate 3 of the Server version. The install took about 15 minutes as opposed to 30 minutes for NT 4.0. The boot cycle seems to be a little quicker as well. So far my in my testing I have not seen a difference in operational performance. The interface well not as elegante as KDE on redhat 5.2 server but the MMC console is a nice touch allowing you to group your adminstrative tools by your responsibility backup, DNS user management.
Granted the bar will be raised on the Hardware requirements Ram, disk space etc. But How many people run their business on a 486 anymore. Not in the Corporate world where I do most of my consulting.
I think performance won't be an issue as much at feature overload with Windows 2000. They are trying to build too much functionality into it so they will miss their release dates again.
Linux will continue to grow in popularity as a stable and robust server platform But NT is not going anywhere by a long shot. Remember Linux is only free if your time is worth nothing.
In a real world environment, a company would hire a sysadmin who knows the platform they are running.....you don't hire an NT person to run a Linux
The benchmark used by Mindcraft spends most of its time responding to simple HTTP requests for static HTML and GIF files. Microsoft implemented a fast path in IIS and NT for serving up such files. This fast path removes process context switching overhead, resulting in faster performance.
The fast path is a system call that writes the contents of a file to a network connection -- it does "cat file to socket" in one system call. I'm no NT kernel expert, but as far as I know this fast path is an undocumented NT kernel extension.
The pragmatic way beat or tie IIS and NT is to add a similar system call to Linux and call it in Apache.
From the point of view of a kernel developer, this "cat file to socket" call is not an attractive kernel extension, because all it does is compose more general purpose primitives into a special purpose bundle. The only advantage of the special purpose bundle is the reduction in context switching overhead. However, that overhead is real, and if we want to win these benchmarks, we have to play the same tricks.
Have you noticed that MSNBC uses NT 3.0? :)
Or the fact that ZDNET uses unix?
Ober
I don't see any rebuttal there. Only a link to the report.
nope they used Intel EtherExpress Pro 100B NICs...
Compaq driver is 2K larger than the floppy? Still using a 720K drive? Last time I D/L the Compaq NIC driver dis for NT it was about 700K.
Since Mindcraft is suppossedly independent (although technically maybe not so) I agree that
reproduction of the results is a very good thing to do. It will do one of the following:
1. Reveal mistakes in the original test
2. Show that the test is not a reproducable, and hence not science.
3. Reveal a shortcoming in the Linux solution that can then be fixed.
One major point here is that it IS possible for Linux based solutions to come up short in certain areas (although I am not convinced that it has happened here). If we bury our heads and just FUD M$ backers ourselves we end up just playing at their level. If a shortcoming does raise its ugly head we need to fix the shortcoming.
>Remember Linux is only free if your time is worth >nothing.
Thats the dumbest thing I have ever seen in print.
From your little windows centric world this may be true, but for many of us it is just the opposite.
Even if the performance part of this test turns out to be poor tuning on the part of the testers, this points out what is a shortcoming. Maybe a tuning guide (if there is already one, I havent checked, disregard this message) with apache is necessary. If there is one, are the critical issues that impact the performance in this test in the guide? Is the guide in an obvious place to a newbie so that it is not overlooked?
Hmmm... that's got me wondering too!
I understand that the Microsoft site (after weaning themselves off Sun some time back) runs a "special" heavily hacked version of IIS,
they wouldn't cheat and use some of that code in this microsoft funded test project, now would they??
Yet another reason for a trustworthy independent testing firm to redo the tests!
It's a bit sad there are no SPECweb96 benchmarks for Linux-based systems, but for a comparison of Apache and IIS, look at
e b96-980713-02887.html
e b96-981103-03152.html
http://www.spec.org/osg/web96/results/res98q3/w
and
http://www.spec.org/osg/web96/results/res98q4/w
I use AMD/Cyrix processor's in the faster machines on the cluster. I believe the performance issues I'm having are with the memory overhead involved with Win98/NT, because if I boot to command prompt only in win98 and run the rc5des client, the number of block processed increases alot, but then I cannot use it for anything else. While with Linux I can run the client in an Xterm with almost no performance hit. You also have to understand, this was not a 3 day test, I have observed this over the course of the last 6 months or so. My averages are based on the each machine processing 24/7, which is what they are suppose to do. Where WinBlows looses is in the reliability area, I spend far too much time rebooting and reinstalling those systems and I put far less stress on those machines than I do on my primary machine. What it boils down to is Linux manages system resources better than Win98/NT and the only way to keep them running reliably is to disable the screen saver, turn the monitor off and not run any programs.
What I would love to see is an NT expert and a Linux expert set up (on the same hardware) the two systems for testing. Then we can know where we really need to work on improving performance for our beloved OS.
Some have suggested, in effect, that we should stop whining about the lack luster performance of Linux, and start developing. Hey, that's great, but shouldn't we first learn what aspects of the Linux kernel TRULY need work, in order to do that?
I'm no OS expert, but I do have a BS in computer science. Just about anybody I graduated with could tell you the impact of context switching on CPU overhead. That's why threads were developed! Maybe this is a good place for us to start......
if you stop and think about it, it makes sense -- since the hardware is custom picked by NT experts to maximise NT's permormance, is tweaked by NT experts.
:-)
firstly, binding network cards to cpus is done on the NT system, but is not possible under linux -- this allows throughput to be increased (shared memory trickery). (could somebody prove me wrong about this)
secondly, the apache benchmark. IIS uses a lot of threads, whereas apache uses separate processes -- and so the instances dont share the same address space -- this incurs a large swapping penaly as the number of processes exceeds approximately 1024 (probably due to the linux setup limiting the RAM to 1024MB -- not sure why, but needs to be looked into).
you should note however, that they DID mention that linux could only see 960 mb of ram whereas NT 4.0 could see 3.0Gb -- note the ratio (i.e. it is approximately the same as the performance difference -- nice one mindcraft
in conclusion:
Had they bought the system configured from someone like Penguin, then they would probably have got better performance in SAMBA. So far as the apache benchmark is concerned -- anybody want to do a native linux threads implementation?
Ever try to install NT with SMP. Holy schnikes, what a nightmare. It kept blue screening on me. After doing a parallel installation to recover and then ripping parts of the kernel out of the sp3 I was finally able to get it to work. It was so long ago I don't remember the exact steps I took. I pretty much use windows only for games now.
I'm sure they just used a configuration which was good for Windows NT Server, i.e. 2-4 processors and enough RAM that it never swaps at all when the data gets in core. (NT is a dog when it swaps)
Linux was designed and optimized on small systems, and NT was designed and optimized on large systems. The legacies show.
Linux's SMP code (I worked on it up through 2.0.30-ish) was made to function correctly and be robust. It works well in user space, but contention in the kernel (i.e. on I/O intensive loads) slows it down.
I'm not sure about Apache, but I bet if you used, say a single processor system with 512 MB RAM (as mentioned, Linux wasn't even using above about 950MB or so) or especially down towards the 64 MB range, then Linux will start beating NT for most things.
Erich Boleyn
Intel PMD architecture
erich@uruk.org
Not only were the tests sponsored by Microsoft, but another report says that NT4 does better than NetWare 5. Another thing, kiddies... if they call it 100Mb ethernet, then how can a server put out more than that (not full duplex) capacity. No. More than one card won't do it. i'm zorrak. just lost the password...
I guess it depends on the need of the business... But I believe any Admin/TSR who knows about both UNIX and NT will tell you NT is annoying. UNIX/Linux needs minimal 3rd party software, while NT relies solely upon it. DNS Manager blows, IIS is the only web server on the net that returns "Sorry, cannot access document. Too busy." (Try logging on to Micro$oft's website when they release a new version of IE or something.) Apache, however, has never had this problem. Neither has any other UNIX based webserver. Also, what hardware was used? NT is a hog and needs alot of muscle just to do the simplest of things, while on the other hand, Linux can handle the same amount of traffic on equal or lesser hardware equivalence. I seriously doubt this is an honest benchmark.
Louis
PS@DAL.net
Not bloody likely, though. It's bad enough to run away with your tail between your legs once.
I'd be delighted if NT clobbered Linux in a fair test. Light a fire under everyone. Don't expect it to happen any time soon.
Really, what we need is an enterprise computing decathalon. Set a limit on total system SRP, or maybe have different categories, flyweight (under 10K$), middleweight (under 20K$), heavyweight (unlimited). Provide a standard set of network firehoses to hook the boxes up to. Have the manufacturers sponsor teams. Of course, the NT licensing model makes price comparisons hard, but lets be charitable and let them enter with a 5 user license.
Of course, crash and you get DQ'd.
Here are some ideas for events:
(1) static web page serving dash.
(2) File transfer clean jerk -- maximum MB/sec served.
(3) File service enduro -- higest number of clients at which you maintain 80% of the peak MB/sec.
(4) Database update derby.
(5) Ecommerce event -- database querying and order processing from pure thin clients using https.
(6) Ultimate Hacking Championship -- no holds barred hacking -- last system left standing wins.
What??? Sysadmin NOT expected to apply patches??? Two things come to my mind immediately: ability to apply [security] patches is a _requirement_ to get a job as a system administrator; every even mediocre experienced sysadmin I know recompiles the Linux kernel to craft it to the hardware _first thing after installation_.
Ever tried to use Solaris out-of-the-box? Ever tried to use WinNT out-of-the-box?
--vt
So, a sysadmin SHOULD be expected to download some 150 MB "service pack", and install it every time they install new hardware or software, but they shouldn't be expected to download a small patch and run a simple patch command? Or, the even simpler command that Shane suggested? They should not benchmark Linux if they don't know what they're doing. Especially not to compare the benchmarks to an NT machine which they DID configure properly. And I have no doubt they downloaded and applied the "service packs" to the half-cooked operating system. Compare out-of-the-box install of NT with an out-of-the-box install with Linux, that's fine. But not only did they compare an improperly configured Linux machine to a well-optimized NT machine, they also did not bother to state that the Linux machine was not configured to take full advantage of the hardware. In fact, they stated just the opposite, that the Linux machine was as optimized as could be.
I searched thru the posts to see if this was mentioned and didn't find anything.
In the posting "will@whistlingfish.net" from Mindcraft made to a few linux newsgroups asking for tuning help he wrote this:
"We're considering using Linux + Apache as a web server. The hardware is a 4-processor 400 MHz (Xeon) server with 1GB of ram, a RAID controller,
and six disks. We have Redhat 5.2 installed and compiled an SMP version of the 2.2.2 Linux kernel. For the web server we used the latest 2.0.3 version of Apache."
Why does it say 1GB of RAM? Didn't the test machine contain 4GB of RAM? If he had said 4GB of RAM couldn't someone have replied to his post telling him about having to patch the kernel to take advantage of the extra memory?
I noticed a few things in their setup for linux that didn't make any sense.
1. they compiled for the 486, while using Xeon processors.
2. they compiled the kernel to have nfs. Why did they do that?
3. it looked liek they only used 1 NIC out of the 4...
Which part is FUD? Queso shows that as of 18:00 UTC today hotmail is running Apache.
>>>>>>>
Not really, because if they were to "benchmark" Linux, they would still have to make NT appear better because Microsoft will bribe them into producing invalid results. If Mindcraft were to release results that show Linux as performing better then NT, Microsoft would certainly release their fury on them. If you have a big foot over you, you can't change direction because of (what seems like) a little mouse.
Spoken like a person whose time is worth nothing.
They weren't necessarily trying to show the performance was better under NT. Their biggest complaint seems to be that they couldn't easily find solutions for their problems. This shows the arrogance of the Linux community. "It's not worth my time to help." Linux is in gross need of support.
They DID post to newsgroups, and no one responds 'til after they post the results of their test. (Time for everyone to slap themselves on the head and say "If I knew then what I know now.") They are exploiting the fact that help for tuning Linux isn't easily found, and no matter how much faster Linux CAN be, it's not easy to do.
Please! The ZDnet numbers you point to were generated by a single 266MHz processor system with 64M of RAM and a single 4G IDE drive.
That's hardly a test of server; in fact, that's about the lowest-end platform available, the sub-$500 desktop.
Using a quad-Xeon with 1G is a much more interesting test.
The response to this is as I feared, the Linux community is just bashing away. I love Linux and use it as my OS of choice. But there are problems with the Linux Kernel when you put it in an Enterprise environment.
1- Linux Kernel uses Syncronous I/O. Which results in wasted time and resources when handling a really large volume of requests. The Kernel needs to be changed to use Asyncronous I/O
2-the Kernel is not completely Reentrant. Until it is, Linux cannot compete in an Enterprise environment
3-Sendfile operation. Linux sendfile API copies the data into buffers that are then handed to the networking code. Solaris UNIX and NT sendfile API sends directly from the cache.
4- the Kernel does not have thread-throttling support.
The honest truth is that until these four problems are fixed, Linux cannot compete with UNIX or NT on the Enterprise level. My hope is that instead of useless bashing, the reset of the Linux community will wake up and address these problems so we can trash NT and UNIX.
-Dethspec
Some of you may remember this from about a year ago. a man who worked in MS's marketing and press-relations group gave some surprising (to some) information with regards to Windows vs. OS/2. At the time, DOS 6.x with Windows 3.1 was battling for marketshare against OS/2 2.x.
His statements included the fact that he did not believe that OS/2 had a single feature that was inferior to Windows and in fact many times was noticeably superior.
His group, unlike IBM's press-relations group, made sure that all software/literature/feature lists were tailored to an individual reviewer. In short, marketing took an active part in the devaluation of an admitted superior product!
Do NOT underestimate the power and influence that Microsoft can demonstrate on the media at large, and consequently the pointy-haired bosses who attempt to make company-wide computing policies.
As a result of this I am now working on an Enterprise Linux/Apache FAQ & Howto.
i se/
s e/)
/. effect so I am looking to collaborate contribute with the above two mirrors and a start of the documents to get the ball rolling.
Mirrors would be greatly appreciated. The site should be available from thurs eve (european time) at http://stacy.flwireless.net/atrevena/LinuxEnterpr
(possible mirror will be
http://www.upsu.plym.ac.uk/~betty/LinuxEnterpri
Submissions and intergration teamwork offers should be sent to betty@area51.upsu.plym.ac.uk
None of the above mentioned sites will cope with a
Aaron. (TheJackal/TeeJay - somebody ate my cookie!)
that's unfortunate.
If this is the case then linux is not well suited to be a server. SMP and RAID support are crucial to a busy internet server. Who in their right mind would put a busy web server connected to a fat connection on a p100?? Hardware-wise you would want multiple processors and an array of drives to maximize throughputs. You would want multiple net cards for the same reason. You would then want an OS that fully support this hardware.
NT has won this particular comparison, regardless of the configuration tweaks to linux's disadvantage. No amount of hand waving or specious whining will change that.
AC
Just a slight correction: SMB predates Microsoft's networking. It was used in Novell's and Lantastic networking long ago. Microsoft did not develop it (at least).
Dan
looser
I think the middle ground would be best. A quad Xeon is a bit pricey, a dual PII 450 might be a reasonable system. However, if you're only using 64MB of RAM, all you're going to show is that NT requires more memory to run efficiently. If you're willing to spend the money to buy NT whats a few hundred dollars to bump it up to 256MB of RAM. If you limit the hardware to what Linux runs well on, I don't think that's a real world test. But that's just my opinion.
Linux isn't free but it is a hell of a lot cheaper than Windows. Given that I work on half a dozen Unix-flavored OS's, all the DOS/Windows flavors, BeOS, and whatever else I hafta I'd say the main thing I like about Unix over GUI OS's is that they do exactly what you tell them right away. In Windows to do something you have to navigate through layers of wizards and menus to get the job done and it takes 10 times as long and often crashes because of some lame ass hidden error you have no way to fix so you spend another three hours trying to bypass the GUI to fix the freakin error. Linux isn't free but if it's based on my time then it sure is a bargain!
Yes, CNN's main web servers run on Solaris. The real video servers run on NT, since Gates paid them to run it.
dublin wrote:
(Don't believe for a second that they really "run Dell on Dell" the way Sun "runs Sun on Sun". So far as I know, Sun and IBM are the only computer companies on the planet capable of running entirely on their own gear. Dell will not be able to replace Tandem and Sun for a very long time yet.)
Hmmm... pass me your crack pipe. tandem is owned by Compaq, which also owns Digital. I think Compaq could easily run Compaq on Compaq. I have a feeling SGI could also run SGI on SGI. HP could run HP on HP without any problems.
That has to be the most ignorant statement I have ever read about Linux. If my time was worth nothing I might entertain the idea of installing Windows NT once a month to keep it stable. I wouldn't mind babysiting the server so that I could be there to reboot it when it crashed. I would be happy to drive to work in the middle of the night to administrate some web server that has crashed since I wouldn't be able to simply telnet into it from home. With all do respect, you are a fucking dipshit if you think NT saves you time.
What you have requested (a level playing field benchmark agreed upon by vendors which any party can submit their own results doing their own tuning) already exists! It's called SPECweb96 (see www.specbench.org). The major conventional system vendors have agreed upon the benchmark, running procedure, and reporting procedure for webserver performance.
There are already 4-way Xeon Windows NT-based results. There are 4-way Xeon Solaris for x86 results. So where are the 4-way Xeon Linux results?!?!
Now unlike what you requested, the results can be submitted on any hardware setup; that doesn't stay constant, making it somewhat tricky to compare results at times. But one-upsmanship on the same hardware is fair game.
Some Linux vendor (Red Hat, Caldera, VA Research, Compaq, etc.) or ambitious user needs to wake up and do these. There would seem to be a substantial payoff in terms of third-party objective SMP credibility for Linux.
--G
http://www.mindcraft.com/whitepapers/response-nove ll-nts4nw5filesvr.html
I agree that not all companies can afford $60K servers, but what happens when the company needs more capacity? The point of the paper, I beleive, is to show that NT scales to enterprise workloads - a place Linux has yet to prove it's abilities.
BTW - where are the Linux TPC and SPECWeb numbers?
The secretary just missed up the paper and they
replaced the word "LINUX" with the word "NT"
hehehehehe
Isn't true ?
CNN is a bigger news site than MSNBC. They run Solaris.
really?
they specified a command line parameter
that possibly does something.
Dont belive any setting until you
verified that it modifies anything.
Ever thougt it may apply on a per processor base?
Ever thougt it may apply only in boot phaes?
Ever thougt it may be overriden by a chipset driver or any other parameter in a higher level
or at least any auto detect?
I hate to tell you all this... I know you're all really big hack-OS fans, but Hotmail runs FreeBSD. Its a known fact that Linux's network code is all jacked up.
-Anon-e-mouse
Look at the comparison at http://macspeedzone.com/4.0
Apple MrC with PowerMathLib:
NUMERIC SORT: Iterations/sec.: 107.164610 Index: 2.769469
STRING SORT: Iterations/sec.: 9.606252 Index: 4.222528
BITFIELD: Iterations/sec.: 3571869546.392807 Index: 612.688667
FP EMULATION: Iterations/sec.: 20.041185 Index: 9.635185
FOURIER: Iterations/sec.: 1602.031493 Index: 1.813977
ASSIGNMENT: Iterations/sec.: 1.510908 Index: 5.756497
IDEA: Iterations/sec.: 462.599684 Index: 7.077719
HUFFMAN: Iterations/sec.: 168.458324 Index: 4.681348
NEURAL NET: Iterations/sec.: 3.053514 Index: 5.166690
LU DECOMPOSITION: Iterations/sec.: 52.623917 Index: 3.106855
This from the bytemark of a PowerMac 604e/180Mhz, compared to a P90 (which would get index=1.0). They both came from a page at Apple, that I linked to slashdot and that seems to have been removed since (no longer available from Altavista). Do some calculations: take the BITFIELDS, iterations per second column ; compute the number of clock cycles taken by iteration considering the 180Mhz frequency of the processor ; conclude.
The "index" column shows that the PPC604e was 600.0 times faster than a P90 (code that you don't run is very fast indeed). This pretty much killed the benchmark. This is why bytemark is plain bogus.
A SIMILAR BYTEMARK BENCH WAS USED BY APPLE TO CLAIM THAT G3 ARE TWICE FASTER THAN PII. Actually they might have taken another compiler to make this claim, but the outrageously better results for the G3 are for the exact same reason, and Apple perfectly knew that its benchmarks were bogus.
Even PC Magazine, a company that almost no one would argue would be biased in favor of Apple vs. Wintel, admits that a G3/400 beats a Dell Dimension XPS T500 Pentium III on tests using an application *widely reported* to be heavily optimized for MMX and SSE! All that with a 20% slower clock speed!
If it is Photoshop benchmark, it is mostly a loop, so not very characteristic of typical applications. Also last time I saw the people added the time of the different filters, which is stupid because it resulted on the slower filters (i.e. unoptimised) being the very preponderant, independantly of how often they are used in real life. Other debatable benchmarks are http://macspeedzone.com/4.0/Wi nvsMacmathematica.html (we don't know which compiler is used, which optimizations, etc.).
Real benchmarks are SPEC CPU benchmarks (see www.spec.org). G3 are about 15% faster than PII at the same clock rate. And of course, a very important point is that all processors aren't available at the same clock speed at the same time. PIII/500 are commercially available that do 22 SPECint. I don't know if I can buy a G3/500 *now* ; and Alpha have had consistently a higher clock than other CPUs.
Another source of SPEC numbers is http://infopad.eecs.berkeley.ed u/CIC/summary/local/ ; if you want to measure the "advocacy effect", have also a look to http://macspeedzone.com/4.0/WinvsM acSPECint.html, which amusingly have lower SPECint numbers for PII than the best published and verified at spec.org (dig spec.org, maybe they took the very first results for PII, or the worse ; there are several different entries with PIIs).
Perhaps you've heard of Project Appleseed? No?d .html
http://exodus.physics.ucla.edu/appleseed/applesee
Imagine that--an independent study of G3 vs. x86 and other processors...hmm...a PII/300 is over 33% slower than a G3/266 (13% slower clock), and over 17% slower than a Rev. A iMac (29% slower clock).
The funny thing is that the cluster of 8 G3/266 would have the same performance of a single 21264/667Mhz for floating point (or 2 21264 for integer code). The 21264 may be more expensive than 8 G3, but it is sure much more easy to debug a mono-processor code :-).
You should have noted the fine prints:
This investigation was initially motivated by the impressive single node performance we achieved on our well-benchmarked suite of plasma particle-in-cell (PIC) simulation codes [4-5] on the Macintosh G3/266, as shown in Table I. This was due in part to the availability of an excellent optimizing Fortran compiler for the Macintosh produced by the Absoft Corporation [6]
and of course:
Computer Push Time Loop Time
Macintosh G3/300: 1750 nsec. 191.1 sec.
One iteration runs in 1.75 microsecond (i.e. 500 cycles), which means that it is really short code and loop intensive (about 100 millions loops), very likely to be optimiseable. This is a specific application, that couldn't be generalized. The annoying thing is that, contrary to SPEC which allows any submitter to choose its own compiler, we don't know how they decided to choose the others' compilers. They are also daily Mac users, they probably know very well how to optimize, and choose the compiler for their Mac.
But all in all, it is a good point for G3. Now if only Apple/Motorola released official SPEC numbers for G3, instead of plain bogus bytemarks...
In response to your comments on the Microsoft TCP stack, you should read RFC 2001 and it's update RFC 2581. Windows clients do benefit when accessing a Web site on NT or other OS's like Solaris. This is not only shown in the Mindcraft report, but is also what a user experiences on the Web. Furthermore, you'll notice that Linux has yet to optimize it's stack to support RFC 2581.
And? Would NT have suffred from the same issue?
You forgot the 4x Intel cards.
Has anyone monitored the traffic on the data lines?
Possibly only one was active due to the standard RH configuration.
Sorry, not the engineers brain prevents fixing
but the possibilities he has to analyze and
configure the system. If i have no insight into
the system at all then i have to reinstall it.
On Linux i have the sources, the compilers,
ways to get more info about any malfunction
situation and finally reliable backup.
Think about that any linux/apache administrator
spends only one day a year on improving the
software... multiply this day by some 100.000 installations and you will get a feeling to
the power!
Read the Fine Print.
... Most developers want browsing technology in the operating system .... "
.... 80% of those surveyed oppose a M$ breakup .... "
Bill paid for this "objective" study !!
You know he's getting sick of all the press every day about Linux kicking NT's ass all over the place. So he ordered this study just like all the other "astroturf" Micro$oft studies you see out there.
"
"
Dream on, losers !!!!
And thanks for the laughs !!!
NT outperforms Linux ?
BWAHAWHAHAWHAHAHAHAHHAHA !!!!!!!!!!!!!!!!!!
Where's the proof?
Where are these benchmarks and don't tell me you are referring to the ZD test done on that desktop machine. I want to see real performacne results like TPC-C, TPC-D and SPECWeb - Oh, I forgot that Linux don't play there. How about other ZD tests on real servers?
How can you make these claims without proof?
Not to mention that there was no pagefile configured on the Linux box (or at least mentioned in the configuration parameters).
There was no mention as to how the four NIC cards where configured (FEC, ALB, seperate networks) and how the routing tables where configured (this can make a big difference)
There was an obvious mis-attempt of not figuring out why Linux would not use the whole 4 Gig of RAM. While NT was left with the whole 4 Gig's
TCP Default Window Size was set to 65535 (was this done on Linux...of course not!)
They disabled every service known to man on the NT box (except what they needed to perform the tests) No mention was made to any services disabled on the RedHat box.
The apache configuration is bogus for the number of hits they planned to use (1000+) They definitely should have given the benefit to turning off some of the mod's that do server parsing (like they did to IIS)
If they used multiple IP addresses for Linux (and multiple IP networks), how did they configure Apache and Samba to effectively use these additional IP addresses. They probably didn't. Therefore, the Linux box was most likely bottlenecked through 1 Nic, while the NT box was allowed to use all 4.
Additionally, the tests did not include an HTTP/1.1 (or HTTP/1.0 keepalive tests) which most peoples web browsers now support.
I would be ashamed to release such glaring mistakes!
ShanMan
I think that testing a modern-day mainframe would have been much more interesting than all this x86 garbage. They didn't even test the pathetic little boring systems like the Sun Enterprise 10000 or the IBM AS/400 architecture.
Remebering winNT is built for networking, not for
ease of use. Linux bunnies bag windows because of
95/98. Leave win95/98 for home users, and NIX & Nt
for networking. Fair competition, pepsi/coke,
wordperfect/word, netscape/ie. NT/(li)(u)NIX.
Quote the original article:
>The honest truth is that until these four
>problems are fixed, Linux cannot compete with
>UNIX or NT on the Enterprise level.
To the first reply to the message:
This was a comment on the enterprise level of nt
versus linux.
There is a net survey, www.netcraft.co.uk is the address. Go to "What's that website running?" and type in the address you'd like to check.
www.hotmail.com is running Apache.
-Benjamin Smith
Remember NT is worth something only if your sleep is worth nothing
Never had a blue screen of death at 3:00AM ???
At least we have a certain view of *what* they think is the worst threat from Linux:
Webserver and Fileserver.
But when ZD themselves used the same benchmarks they go different results...7 506,00.html
http://www.zdnet.com/sr/stories/issue/0,4537,38
Yes!
Yes, and they didn't even have the Netware box tuned properly. If they couldn't get a Netware box tuned properly I doubt they could tune a Linux box properly either.
>Well I agree with 'im, and my time is worth
>plenty to those that fill my bank account.
>I work on NT. I use Linux for my home network.
>And my home clients are the most demanding.
Obviously you don't use NT for any real work.
From my experience if your time is worth anything (like mine is) NT costs alot. My belief from my experience of Linux/Unix is that it would be far cheaper for a company to use them than NT.
Why should NT be crippled by the inabilities of Linux? Standard NT supports 4GB of RAM. NT 5 will support 64GB.
Maybe this is why Linux can't post TPC results - they'd suck.
Name: msnbc.com
Addresses: 207.46.150.209, 207.46.150.213, 207.46.148.249, 207.46.148.250
207.46.148.251, 207.46.150.201, 207.46.150.205
Aliases: www.msnbc.com
Name: slashdot.org
Address: 206.170.14.75
Aliases: www.slashdot.org
7 MSNBC servers to 1 slashdot server.. Seeing as each of those msnbc servers are probably 4 CPU machines, and slashdot is probably a single p300-ish machine, no wonder it outperforms slashdot.
IIS is nice if you have an overblown budget and want to spend all your money. Otherwise, it's pretty much useless.
new kernels are maybe around 15meg downloads...
I read Mindcraft's "rebuttal" (if such a poorly written document can be considered a rebuttal) of Novell's criticism of an NT vs. Netware test performed by Mindcraft. I found it to be full of grammatical errors, including such basic mistakes as incorrect subject-verb agreement. In any case, the rebuttal is filled with worthless double talk and support for their claims is absent. As far as I am concerned, this company can not be taken seriously, as they (as a "professional" independent lab) cannot even afford to hire someone to proofread their releases to the public. I also read their white paper regarding the NT vs. Linux test that they did, and found it to be extremely biased. Why anyone would set up Samba with widelinks=no is beyond me. That imposes a terrible bottleneck on the performance of Linux as a file server. In any case, I found the Mindcraft studies to be biased toward NT and will not take any future study by them to be true, objective information.
so there.
Shows you how nice it is to have a reliable, supported OS.
"A statistician uses statistics in the same way a drunk uses a lamp post - more for support and less for illumination."
Note the hardware used in the test: A couple of quad-processor machines with RAID. This sort of configuration is more on the fringe of Linux's abilities, while NT is supposed to work at its peak on a 4-processor system, and I'm sure has much more mature RAID support.
I'll bet if the test were repeated on a couple of single-processor boxes with standard IDE disks, the results would be very different.
I'd walk away from this test with the following conclusion: Linux needs more tuning for higher-end hardware.
Of course, note the spin of the article: If you don't read closely, it looks like NT is 2.5 times faster than Linux in some sort of overall sense.
Mindcraft's 'credibility' is blown (if it had any to start with) as any company that seeks to do nothing twist tests to meet the clients satisfaction can not be considered as any better than straight MS marketing/FUD.
I note that the tests & machine vary widely from previous Netware vs NT tests - why?
No mention of relative cost is made which is strange as cost/performance is a rather important factor (how much is NT with 140+ client licenses anyway?)
Why take a machine with 1GB of RAM - is this typical of the average PC server?
MS are simply hoping that media will simple report that 'NT is 3.5 times faster than Linux' as they assume (rightly?) that is all corporates will remember. The only answer is to ignore the Mindcraft study and keep publishing (carefully selected of course) benchmarks showing Linux speed. Thats all that counts in the end.
here's the reply they got from comp.os.linux.networking:
I hope it's not too late to change your hardware, because your box is a
complete waste of money. SMP gives you *nothing* with regards to web
serving, and it makes your OS flaky as all hell. The RAM is nice, but the
processor speed is overkill and having 4 of them is just plain wasteful. The
network card would saturate completely before you even came remotely close to
using up the resources of even a single P2 200Mhz.
They claim that NT is better than SOLARIS and NETWARE. Just go to http://www.mindcraft.com/. Cut and paste you lazy bastards!
They are just killing their own credibility, not that of Linux.
The apache they used had mmap. I was told that could cause slowness... is it possible that was the cause for apache tests.
I do know Linux supports the slowstart standard on tcp/ip. 2.2.x has this, does NT? If not, is this what does it?
Trever
And what the hell is this?
Set OPTIM = "-04 -m486" before compiling
Set EXTRA_CFLAGS=-DHARD_SERVER_LIMIT=500
They set a HRD LIMIT of 500 connections and then mourn Apache crapping out at 1000 connections? Puhleeze!
Wankers indeed!
I think there is some kind of misunderstanding somewhere.... shtml .... 2,5 times and 3,7 times.. (including lotsa graphs etc)
http://www.gcs.bc.ca/bem/editorials/nts4rhlinux
this page says exactly the reverse story, i.e. RH5.2 is faster than NT4 v/sp4
I've read through a lot of postings to this case, and an overwhelming amount whining about conspirations and buy-offs and misconfigurations and what-have-you. Maybe. Then again..maybe not. This is what psychologists refer to as "Denial". Let's stop whining and start coping and start DEVELOPING.
"Shut the fuck up! NEXT!"
-Coward
Gee, this stuff looks so darned familiar. Back in the OS/2 days, there were any number of 'MS inspired' benchmarks like this showing NT (or other MS product) beating OS/2 in benchmarks. Typically, the situation showed a hand-tuned MS system with carefully selected hardware, and a hand de-tuned OS/2 system.
Deja Vu, all over again.
As someone who works in the real world, I have indeed found NT to perform better than Linux as a SMB file server...until you put a load on it.
.02
When NT 3.51 first came out, I was a big proponent (vs Novell), due primarily to costs. As an educational institution, we couldn't afford Novell, it was that simple. However, over the years, I have learned that NT just can't handle it. As soon as you throw 30 workstations at an NT Server, it starts to grind to a halt.
My real world tests show that NT is indeed faster than Linux at first, but soon starts to bog down to a point that the only real option is to start adding more servers.
This is what Microsloth doesn't like you to know about. Only when it's too late and you are forced to buy more servers and clients, are you awakened to the TRUTH.
One last comment.
Nobody has mentioned cost analysis and ROI in any of these benchmark studies. For an enterprise/institution, what is the total cost of ownership breakdown between Linux and NT?
I'd be willing to bet that if/when corporate america figures out that they could save tons of dough and actually increase the usage of their servers, Microsloth will be on it's ass.
Just my
This seems to be a variant on the wishful thinking "ignore them and all they say and the problem will go away". It does not have to be like you describe. Allow me a simple thought experiment to demonstrate my point:
Imagine two systems. System (A) is a low end server like you describe, say a Pentium II 200 with 64 mb ram and a 10 gig drive. System (B) is one of the systems they used in this test, or any given quad processor xeon with 5 drives and 4 ethernet cards.
Strip these system of any OS differences. Which one has the better theoretical performance? Even counting hardware designed for certain OS features, system (B) is the clear winner on hardware alone.
Imagine that (A) running linux outperforms (B) running NT. What an amazing feat that would be, but just imagine that linux is that good and NT is that bad. Now put linux on system (B). You see where this is going. Even allowing for M$ manipulated "independant" testing agencies to tweak out the performance from linux and tweak in extra performance for NT, there shouldn't be much of a contest. Linux must absolutely shine when it is given the hardware to do so.
My point now is that linux has been dealt a credibility blow and the original post in this thread is spot on. Linux must have beefier SMP support and better RAID support as well. And these items must be available "out of the box", even if only in certain specialized distributions.
Let's take this as a challenge and run with it.
Dan
Fished out of dejanews -- the Mindcraft folks used the pseudonym 'will@whistlingfish.net'
(If this was posted earlier, I didn't see it...)
Can anybody here respond to this?
Hi Everybody,
We're considering using Linux + Apache as a web server. The hardware
is a 4-processor 400 MHz (Xeon) server with 1GB of ram, a RAID controller,
and six disks. We have Redhat 5.2 installed and compiled an SMP version
of the 2.2.2 Linux kernel. For the web server we used the latest 2.0.3
version of Apache.
The scenario: we're bangin' on this web server with a bunch of clients
to try and get a handle on its capacity. Simple static HTML requests,
no heavy CGI yet. My Apache server is tuned up, MaxClients is 460.
I recompiled with HARD_SERVER_LIMIT set to 500. Limit on number of
processes is 512, imit on file descriptors is 1024.
The problem: the server performs well, delivering in excess of 1300
HTTP GET requests per second. But then performance drops WAAAY
off, like down to 70 connections per second. We're not swapping,
the network isn't saturated (4 x 100Mbit nets), disks are hardly used,
but the system is just crawling. If it were saturated then performance
should level off, not drop like this. Neither vmstat nor top show
anything unusual. No error messages in the web server. Its puzzling.
Any ideas? Any tips, suggestions, or pointers would be appreciated.
Thanks!
In the article it says that they used _Samba_ 2.0.3.. Maybe somebody got confused and put the wrong information in the posting?
(i'm not familiar with the latest releases of samba and apache, so don't sue me on this..)
Bleeding Edge Magazine did an extensive "test" today (which was sponsored by Red Hat Software and VA Research, by the way) which proves that Linux is faster than NT than Web and file serving. PHB's, watch out.
If you think it's that important, feel free to fix it, making memory size a Makefile or config option and send a patch to Linus.
So, you really mean that you should use pthreads. I could see that. But pthreads aren't nearly so platform-independant as fork() is.
Based on your post ... in fact, based on your subject, I'd say that every one of the Apache Group's programmers are better programmers than you.
Let me count the ways:
- You claim that ``libc6'' (posix) threads are ``real threads'' implying that fork() does not make ``real threads''. Procesess are definately real threads, except they have more overhead. In fact, the terminology for pthread and pthread-type threading architectures on unix-like machines (such as pthreads or solaris threads on solaris) is ``lightweight threads'' or ``lightweight processes.'' They are just like other processes except that they don't have the overhead of things like seperate page table, process id blocks, memory spaces, etc. Both lightweight (pthread) and heavyweight (process) get swapped with the kernel scheduler, however. And you should also remember that pages are only copied on write, so the actual memory footprint difference isn't all that great.
- You say that limits on the number of threads is a bad thing. Of course this shows you lack both programming and system administration knowledge. If there was no cap on the number of threads, process or otherwise, it would be trivial to make a denial-of-service attack that would render the machine useless. And setting an arbitrary cap in the code is rather silly, as some architectures allow for more copies of the server running at the same time.
- You say that having lots of options is a ``DUMB'' way to do things. I'd say the opposite. Having software choose your options for you makes an idiot out of the user. Configurability of almost everything is one of the greatest strengths of UNIX software. While YOU may not think that some option is necessary, it may make a huge difference in functionality for someone else. Apache has good defaults for the typical small web server. It has configurability if you need high performance or very-low footprint.
Tell you what, rather than complain about the Apache Group's informed decision to NOT switch to pthreads, why don't you try to port to pthreads yourself? If the benefits are as great as you say, then I'm sure that your product would be a great asset to the community. I think that the Apache group has done a great job, that you overestimate the benefits of pthreads, and that the loss of portability by switching Apache to pthreads would be heartbreaking.-- Erich
Slashdot reader since 1997
slashdot broke my sig
On top of inetd, what if they used tcpd as well? Or the "KeepAlive Off" in apache's config? They don't specify the ServerType parameter, but in my httpd.conf, the comments say the default is Off. Talk about slowdown, one request per connection each having to be thread through inetd (and possibly tcpd)...
One thing that's bothered me through all hype about Linux sucking, these sort of "studies," etc: why are they always run by people who have no clue about the things they want to portray that they are experts on? Sure there isn't much to NT, click some Next buttons through wizards, and voila. So they apply that same sort of mentality to Linux, either taking a bare RedHat (or other distribution), or minimal customization. (The recompiling kernel causes you to muck up the entire system beyond recognition bit, my guess is they have no clue about bootable floppies, configuring LILO to have two kernel images for fallback, etc).
What about the 960MB memory thing? Just a matter of telling LILO append="mem=1024M" ? I know it freaked when I put in 96MB the first time, only seeing 64MB.
As others have said, the posts the the newsgroup contained some major flaws, not enough details, etc. That certainly would turn off many potential replies.
Microsoft sponsoring them? Wouldn't their credibilty be higher if sponsors were NOT the manufacturers of the products they are testing? To me that's a major problem. For respect, a study should be balanced and unbiased.
In conclusion, they are lunatics. Plain, simple.
What are they supposed to do, take the money MS gave them for the study and spend it all on a mediocre sysadmin? Where's the fun in that? Look at what they have accomplished though. They get money from MS, they say NT kicks ass, MS is happy and may continue doing business with them in the future, their summary will be plastered everywhere (boy, that NT graph is higher, it must be tons better than this Linux thing), they get more and more attention. Very few PHBs these studies target read the details, know what parameters to Samba/Apache do, etc. And all these organizations continue spreading FUD studies...
Ah what a world we live in.
Apache collapsed after 250 threads on a Quad 400Mhz Xeon? Something is definately screwy there..
I have average close to 60-70 Apache threads running as a regular load on Pentium-120 with 64megs of ram without any problems. Most of those are database-generated, rather than plain file GETs. Someone has been either drinking or got paid some dough..
Because the bug was fixed in 2.2.3 which was out when they did the tests.
> Used 1024 MB of RAM (set maxmem=1024 in boot.ini)
Maybe it's just me, but the fact that they went to the trouble of editing the boot.ini but not the lilo.conf is suspicious. Is mem=1024M really that hard? I'm quite certain the feature is documented.
If a Linux kernel can only see 970 MB RAM, it's been misconfigured. To make it see more, there's a source file that has to be edited - but why isn't it an option selectable from 'make config'? SMP is... maybe this'll motivate the change.
> The reason any truely seasoned IT professional
> scoffs at this document is because of the
> source.
True, but I notice you qualified that observation with "seasoned." Alas, these same seasoned systems folks aren't making many purchasing decisions and are kept in the back room where their views on reality don't embarass the suits. (speaking from experience here...)
I've encountered too many people mentally conditioned by Pravda who will discount any all other studies - no matter how technically solid - in favour of the ones - no matter how technically soft - that support their prejudices. (Sadly, MS apparatchiks are not the only ones guilty of this).
Oh man, I feel so enlightened by their wonderful studies!
... I need to modify/recompile rdist to use ssh instead of rsh on some intranet servers. It was so easy on Linux and Solaris so it must be a snap on NT eh? I wanna do this on NT so I can turf all those silly Sun 450 boxen? They're obviously useless as servers.
I mean wow! NT/Compaq blows away a UE450 worse than it blows away Linux?!? I guess I should stop wasting my time with this old hat Unix junk and get with the program and join the winning team! You can be sure that Real Soon Now NT will be far more reliable and scalable than any Unix system and I will be straight out of a job if I don't embrace the New Technology and join the marching ranks of brave new world techies towards progress and bliss.
Now
I see M$ is still up to their old tricks. Oh well. It'll be nice in a few years when they're not even around anymore.
There is one question: why Linux should have that? What study (other than very NT-biased one, made by Russinovich) are those requirements based on?
Contrary to the popular belief, there indeed is no God.
----
Every year during my review, I just pray the words "slashdot.org" aren't mentioned.
Really; the Winbox had most of its services shut off, while the Linbox was running SMB, NFS, etc. My guess is that they were probably hitting those other services while they were taking the numbers.
Besides, this runs contrary to every other (non-MS paid-for) study I've seen. Mayhaps someone should do some independent verification. Be sure check if the Windows numbers were a "demo".
Hey, they lied to Justice; why wouldn't they lie to us?
----
Every year during my review, I just pray the words "slashdot.org" aren't mentioned.
Well, 2.2.2 *was* listed as a stable kernel after all, so why would they have any reason to expect otherwise?
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
Here is an SMB test on a large machine.
In general there are some areas where Linux lags NT. IIS, for example, outperforms Linux on static page displays because it has a page cache and does not have to always cross the user-kernel boundary to fetch the page (system calls DO have a cost, even though 2.2 sped up the open() call considerably with the dcache). And it may very well be that a well-tuned ultra-high-end NT machine will beat a well-tuned Linux machine at file serving, given that NT will support the full 4gb of memory while Linux only supports 2gb of memory. But this "test" was not such a test -- it compared a well-tuned NT machine against a totally untuned Linux machine.
And of course I'll point out that on tests on more modest hardware, like this, Linux blows away NT handily. To be fair, that Smart Reseller test was just as biased in its own way as this joke test we're talking about... Smart Reseller chose a machine that's too small for NT to comfortably stretch its legs, albeit that the machine they chose is rather typical of small office web servers.
-Eric
Send mail here if you want to reach me.
Hard to get newsgroup support for that sort of stuff? You bet!
I know that I had no incentive to go dig this "will @ whistlingfish" out of his hole. I couldn't make heads nor tails of that posting when it was new, and there's too many other postings to reply to where people actually give useful information about their problem for me to bother with something like that.
-- Eric
Send mail here if you want to reach me.
It's a little late to sponsor such a contest. It takes a week to ship something via motor freight from either coast to Chicago. Believe me, a 200-pound server is *NOT* shipped via Federal Express Overnight Air!
-- Eric
Send mail here if you want to reach me.
1) As someone pointed out to me via EMAIL, Apache actually opens most of those file handles prior to forking. Thus most of them are shared between the various Apache processes. In actuality, you'll eat up at least two file handles for each Apache process.
2) The 2.2 kernel defaults to 4096 file handles, as vs. the 1024 default for the 2.0 kernel, so it's unlikely that he was running out of file handles.
Still, obviously he did something wrong, because Apache usually does not collapse like that. It simply degrades gracefully, assuming max_clients is set so that you don't thrash the machine to death (and his message says he wasn't thrashing). See what happened when the Slashdot Effect hit the Linux Counter... once he brought down his max, it simply got slow but kept chugging out the requests. Puzzling. Without access to the server logs and httpd.conf files, it's unlikely we'll ever know what or how he did it, though.
00 Eric
Send mail here if you want to reach me.
I agree, it is hard to believe.
And: These people are *LIARS*. They say the posted messages asking for help on the Linux newsgroups. There are *NO* messages from the mindcraft domain anywhere on the Linux newsgroups. So I did DejaNews searches of "performance tuning", "performance tune", "kernel tuning", "kernel 2.2 tuning", between January 1 1999 and today, and examined the results to see if there were any messages that may have been by MindCraft researchers (i.e., that were referring to performance problems with a large-memory machine). There were *NONE*. Zero. Zilch. Which means that if they did ask any performance tuning questions, they did not use those words in the message.
Anyhow: VA Research already loaned a quad-processor Xeon machine to PC-Week and it blew away NT 4.0 in their SAMBA benchmarks. VA Research's quad-processor Xeon machine is the same machine that we sell, and the same machine that Penguin Computing sells (we all get them from Intel, and then dress them slightly differently once we get them, e.g. VA Research uses a Mylex RAID card while we and Penguin use ICP-Vortex RAID cards). So we already have the benchmark that shows that their SAMBA benchmark is full of ****. But that's not going to matter to pointy-haired bosses because they recognize only those reports and studies that say what they want to hear.
Am I steamed? You bet! I *HATE* liars!
-- Eric
Send mail here if you want to reach me.
that they had the Apache config set to allow up to 127 servers, and they had not raised file_max from 1024 to something decent. Do the math. Each Apache server has 8 file handles open just sitting there doing nothing. If it then goes to open a file to serve it, and file_max has been exceeded, guess what happens? Yep, Apache collapses!
-- Eric
Send mail here if you want to reach me.
Thanks. I just checked that out. It does appear that they asked a single question about Apache performance. I remember seeing that posting myself and blowing it off because there wasn't enough info to tell him anything and I didn't feel like going into the give-and-take to get enough info to do something. (I do enough of that supporting my own customers!). Now, in hindsight, knowing what he did not do to Linux, the answer is obvious: he was running out of file handles. Do the math. An inactive Apache server has 8 file handles open. 127 servers max * 8 = 1016. Default file_max is 1024 for Linux, of which 150 or so are usually open while the system is at a rest. Apache could not bind a socket to a file handle for incoming connections because there were no file handles. So Apache was basically deadlocked, waiting for file handles to come free so it could accept() the socket but it was already holding all the file handles!
If there had been questions about general tuning of such a large system, that would have solved the problem because someone would have remembered about file_max. But one cryptic query that didn't give enough information to get help does not an honest effort make.
Anyhow: I guess I have to post a partial retraction. They did post a *SINGLE* query to the net.
-- Eric
Send mail here if you want to reach me.
Novell reacts to Mindcraft Benchmarks of Novell 5 vs NT4.
e svr.html
http://www.novell.com/adv antage/nw5/nw5-mindcraftcheck.html
Mindcraft admits that Microsoft commissioned the original report.
http://www.mind craft.com/whitepapers/rebuttal-summary-nts4nw5fil
Do we see a trend here?
Codifex Maximus ~ In search of... a shorter sig.
Sorry, whats the full URL? I can't seem to find it on their site.
Sig (appended to the end of comments you post, 120 chars)
I've heard that NT has a 300MB file cache limit that is supposedly well documented. PC Week did a Samba benchmark with NT vs. Linux and their conclusions were so bad (for NT) they didn't even publish them. This was because NT would not use over 300MB for its file cache while Linux would use 900+MB.
How did Mindcraft get around this, or was it actually documented in their report? Heh, I guess I should probably read the entire thing first huh? Naa!
Also, after some research on why Apache performance dropped considerably at one point, it does look like they hit the 1024 file descriptor limit. Alan Cox has a patch for 2.2.x that brings this up to 10,000+ and theoretically millions. Check the recent Kernel Traffic mailing list for details. Did they do any research at all for this report, I mean come on!
Sig (appended to the end of comments you post, 120 chars)
Yup, and that extra cash you could probably hire a much better Linux sysadmin ;)
Sig (appended to the end of comments you post, 120 chars)
Yup, and that extra cache you could probably hire a much better Linux sysadmin ;)
Sig (appended to the end of comments you post, 120 chars)
What we really need is a comparisions of two systems with identical hardware tuned and configured by memebrs of their respective camp.
ie a Windows NT 4.0 server running IIS setup by microsoft employees against a Linux 2.2.2 server running Apache setup by Redhat? people. But we all know that MS would never allow such an unbias test to occur under their rule of FUD.
Posted by the order of His Majesty:
I've heard that NT isn't anywhere near ready for Gigabit anything. Reading in NetworkWorld about some network show, they mentioned how they talked with several gigabit ethernet vendors who where very unimpressed with WinNT's throughput - and commented that NT had to be specially modified to sustain 400KBytes/sec.
I haven't ever tryed NT with gigabit ethernet, but it doesn't surprise me...
Posted by mithalas:
960meg 4 CPU's. Hmm. I want to know why any web server is going to need (not to say that you don't want it) that much ram, or even 4 processors. If we all set back and think about if you really need that much for your server to perform then you probly are running a pretty crappy os. Granted NT has it's place and GNU/Linux (no flames please) has it's. Personally I would like to see a more down to earth server tested. How many web sites out there run 4 CPU 1gig servers. I say take a poll (running x86 based CPU's) of 50 of the largest site, 50 of the middle sites, 50 of the smallest sites and get what they run on (i.e. HW). They come up with an average take that average and put it together then install your os's of choice with experts from all sides to configure the os's. Then run the tests again. You'll probly see that you end up with Linux winning. From personal experience NT is no easier to configure than Linux maybe get installed initially but not configured. An untrained monkey could install it, but you better have time and experience to make it work more that 2 weeks for you with out a crash. When your lucky you get a whole month.
Mithalas
MCSE
Posted by mithalas:
Please show me on an official test NON BIAS that with that entry nt will only use 1gb. Plus you should checkout what they did with NW5 and the interesting ways of configuring hardware.
Mithalas
MCSE
Posted by llogiq:
Is this topic really worth to write dozens of postings about? (Guess why we all are posting here...) Well, even if NT was twice faster (which it is not), would anyone matter?
linux just started to gain sympathy even in business - some (who said they would not risk using a hacker system some months ago) already say they can't afford proprietary systems like M$ NT or else - now.
Another thing is that it's nearly impossible for a bunch of well-paid M$-coders to improve a server app like the thousands of (free) linux coders do.
Posted by Nr9:
i think mindcraft is part of ms, all they do is advocate ms products
Posted by Nr9:
blah, www.hotmail.com is freebsd but its other servers are solaris
Posted by laziz:
I just read the whitepaper, and I see a few possible criticisms:
1. The performance numbers suspiciously suggest that the NT box was using
more of the 4 available NICs than the linux box.
2. They didn't optimize the kernel for PII compilation.
3. It doesn't look like they created a swap partition for linux-- could
this cause the VM subsystem to freak out?
4. Contextual mistake-- they're using (what appears to be) beta drivers for the RAID card. I'd like to see the numbers for the same box without
the raid card, just two disks hanging off an ultra-scsi adapter. It seems
that they picked a platform with as many linux-compatibility issues as
they could find.
Just my $0.02
Posted by wclark:
:)]
Nobody that's run both will dispute the fact that FreeBSD's networking code is and always has been faster than Linux's. That's _NOT_ to say that Linux doesn't outperform FreeBSD in most every other area, or that BSD will always be the fastest. I hate Slowlaris with a passion, but that doesn't stop me from buying Sparcs for databases. You go with what works best in each situation [but in the mean time, I still cross my fingers and hope Linux matures and proves itself as a database platform, so I never have to network troubleshoot a box with no `traceroute` and a hobbled `ping` again
Posted by The Chicken of Darkness:
Let's see here... they were using a ZD program to test SMB performance? To a certain degree NT would have an edge, given MS made the SMB protocol and since it is a Ziff-Davis program. The headline doesn't even specify "fileserver". I'm sure Apache would shread NT as a netserver, performance wise and in up-time.
Posted by LOTHAR, of the Hill People:
c es:
..."
I supect that the people conducting the test were not proficient linux users/administrators. The Linux installation followed defalut settings except for Kernal automounting. The fact that
"NFS file system support = yes "
makes me wonder how the drives were partitioned (RAID configuration as well)
The test also mentioned
"The Linux kernel limited itself to use only 960 MB of RAM"
Which is a subject dicussed here last week
The NT installation was not default, the Registry was directly changed.
"Server set to maximize throughput for file sharing"
"Set registry entries: HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Servi
\NDIS\Parameters\ProcessorAffinityMask=0
Tcpip\Parameters\Tcpwindowsize = 65535"
"Used the affinity tool
I do not consider the test valid.
The NT installation was tuned (if even slightly)
The Linux installation was not.
We need o third party with a though understanding of both OS's to administer an accurate test.
Posted by the order of His Majesty:
Not regarding the web server, but I noticed that they set NT's pagefile to 1GB and didn't mention Linux's swap configuration at all.
Taking everything that they did to misconfigure Linux it doesn't surprise me that it doesn't perform spectacularly, it's like turning off L1 and L2 cache, turning off shadow ram, and setting 4 wait states to a PentiumPro 200 - it turns into a 386!
Stuff like this ought to let people see what Microsoft's game plan really is (assuming they even have one, after reading the Halloween docs (: )
Does anyone know if Apache needs to be recompiled to support SMP or if there are SMP optimizations that can be added to it readily? If MS server had such optimizations and Apache didn't that might cause a significant performance to the Apache/Linux server. Same questions about SAMBA.
Later,
e.
P.S. There is a lot of MS thrashing in this thread. Let's please try to stick to exchanging information and ideas. After that we can say they suck.
Given their network configuration surely they would have to have all the network cards running in order to get more than 1/4 of their machines running....and the performance figures would seem to indicate that they did have it all running.
I don't think these people are stupid...
Ah yes...but they were running Win95/98... if you machines crash twice a day then you aint going to get many blocks done.
Had they been running NT4 workstation or Linux it would be a pretty cool setup. My employer has a similarly large number of PCs running NT... and they spend all of their time in the idle loop.arrggh!
That's Disraeli, hoss. Not Churchill.
Remember that what's inside of you doesn't matter because nobody can see it.
HTTP Error 403
403.9 Access Forbidden: Too many users are connected
This error can be caused if the Web server is busy and cannot process your request due to heavy traffic. Please try to connect again later.
Please contact the Web server's administrator if the problem persists.
------------
HTTP/1.1 200 OK
Server: Microsoft-IIS/4.0
Date: Wed, 14 Apr 1999 03:29:45 GMT
------------
Here's a nickel kid, buy yourself a real operating system.
HTTP/1.1 200 OK
Server: Microsoft-IIS/4.0
Date: Wed, 14 Apr 1999 03:33:38 GMT
Content-Type: text/html
Set-Cookie: ASPSESSIONIDFFFFCDDH=PIMBKNKALGKHCHKFMBAGPACB; path=/
Cache-control: private
Okay, this is rubbish. Sponsored by MS, eh. Weeelll, then it must be true! Here are some other points of view:
1. This article in InfoWorld about MS's own concern about NT server system crashes, along with a few choice quotes from other consultants.
2. This article by Novell pointing out Mindcraft's lack of, uh, a reality check.
3. This survey by Netcraft showing Apache whopping the sh*t out of NT/IIS.
Preaching to the choir, I know, but I just had to do it. I just wish we had a central, big-name pr channel to counter the MS FUD. Any ideas?
"shop smart:shop s-mart" ash
From their website: The Benchmark for Web Servers WebStone 2.5 We've just released WebStone 2.5. Both executable and source code are available for free and will stay that way. The highlights of the new release are: It's much more reliable. Tests are reproducible. The previous version of WebStone used a time of day random seed. Now it can optionally use a fixed random seed. WebStone 2.5 comes with a new license designed to encourage you to contribute enhancements while protecting you and your company from liability. There is an optional new dynamic workload that simulates ad rotation. WebStone 2.5 results can be compared to WebStone 2.01 results (unless you use the new dynamic workload). What have we accomplished here? Ooh, we made yet another friggin' license to check for freedom, and admitted that... the last version didn't support "reproducible results"? *chuckle* Yep, that sounds like 'The Benchmark for Web Servers' to me. Maybe I'll try it on their web server...
pb Reply or e-mail; don't vaguely moderate.
I'd like to see them contact someone who knows how to configure a Linux box before saying they ran benchmarks showing something. They claim to have contacted Red Hat and a couple of newsgroups for help, but they don't apparently have that many UNIX gurus on staff...
Maybe if they'd, say, bought a box from VA Research for this purpose, or asked them for help, or contacted the Samba team or the apache team...
Heck, if I had too much money to toss around, I'd be happy to run benchmarks like this, and refute their claims. Linux ran wonderfully on my P133, and NT4 (any revision, later ones are worse) runs horribly on it. I've never seen NT do anything besides eat memory and watch applications crash.
pb Reply or e-mail; don't vaguely moderate.
Aww man, I'm sorry, guys. I didn't realize the "PRE" tag wasn't supported. I guess I'll just keep posting in text. :)
From their website:
The Benchmark for Web Servers
WebStone 2.5
We've just released WebStone 2.5. Both executable and source code are available for free and will
stay that way. The highlights of the new release are:
It's much more reliable.
Tests are reproducible. The previous version of WebStone used a time of day random
seed. Now it can optionally use a fixed random seed.
WebStone 2.5 comes with a new license designed to encourage you to contribute
enhancements while protecting you and your company from liability.
There is an optional new dynamic workload that simulates ad rotation.
WebStone 2.5 results can be compared to WebStone 2.01 results (unless you use the
new dynamic workload).
What have we accomplished here? Ooh, we made yet another friggin' license to check for freedom, and admitted that... the last version didn't support "reproducible results"? *chuckle* Yep, that sounds like 'The Benchmark for Web Servers' to me. Maybe I'll try it on their web server...
pb Reply or e-mail; don't vaguely moderate.
and found this gem about what happens when NT needs tuning. *chuckle*
I guess tuning makes all the difference.
pb Reply or e-mail; don't vaguely moderate.
:) Yep.
From Whetstone to Dhrystone to Whebstone.
That's really horrible.
I guess if we ever make "The Ultimate Pretentious Open-Source Buzzword-Compliant Web-Server Benchmark", we've got a name for it.
pb Reply or e-mail; don't vaguely moderate.
I have written many papers in LaTex
and I find it much easier to work with
and more portable than say Word or WordPerfect
Now for something that needs to be done quickly
with the least amount of effort
(i.e Learning a new language)
you might use Word or equivalent
But for a large job like a few hundred word thesis
or a book, I would prefer to use LateX
Because it makes my job of typesetting much easier
If they had availed themselves of a Linux expert (or gotten Linux pre-installed by a good VAR), they would have tuned the kernel to at least use 2G of RAM. All you have to do is change __PAGE_OFFSET to 0x80000000.
Simon Cooke is an MS employee. He regularly posts on the MacCentral forums so he is no stranger to me (unfortunately).
Hi Simon! Its Tak!
It's far easier to forgive your enemy after you get even with him.
It's far easier to forgive your enemy after you get even with him.
enough said
_damnit_
It's my job to freeze you. -- Logan's Run
In my travels, I've come across a page that illustrates that. It's here on thttpd's site It's a bit dated, but it does illustrate Zeus' (and thttpd's) speed.
Compare and contrast:
NT 4.0 is 2.5 times faster than Linux as a File Server and 3.7 times faster as a Web Server
with ZDNet's findings on the subject of the same benchmarks of Linux vs NT benches.
The same thing could have happened if smp were "Accidentally" left out of the linux machine's kernel.
I doubt that any amount of "tuning" would generate this type of difference. With apache, perhaps realtime ip resolution would do it, but I don't see how that would figure in with Samba.
An older PC Week article had a showdown between several web servers. On CGI, Apache on RH Linux beat everyone but Apache on a dual processor Solaris box. The problem was, the PC Week guys couldn't figure out how to compile the kernel for multi-processors (actually useful for CGI rather than plain static pages). Static pages, the RH/Apache solution performed almost identical to IIS.
--
Ben Kosse
Remember Ed Curry!
You paid $170 for a domain name to host on a Win98 box. If that's any sign of the Microsoft groveling butt kissing that computer science has become, I'd rather work as a minimum wage biology lab tech forever.
What are these news items about server downtimes and McAfree virus scan? How the hell do you give a virus to an operating system? Do tell.
Time to give your Win98 box a few reboots.
10000!!! The default on my Debian system is 30.
I am not an Apache guru, but this dubious parameter setting may be part of their fiendish systematic detuning of their Linux system.
By setting this value so high, they exacerbate memory leaks. This might be why their performance collapsed after a while.
-- Mike Greaves
Since Slashdot is running a heavy-duty Linux webserver on an SMP machine, the httpd.conf, etc. from this site might make for an interesting comparison.
MaxRequestsPerChild 10000 indeed!
-- Mike Greaves
So I guess Microsoft should be migrating all of Hotmail's services to NT servers now, right? Well? I'm waiting...
That's what I thought.
My point was simply that if NT is really that much faster than Linux as a file server and a web server, then there are only two things they could be saying by not running Hotmail exclusively on their own OS:
1) The combination of FreeBSD and Solaris is far superior to both NT and Linux, or
2) NT is superior to FreeBSD, Solaris, and Linux, yet we still won't use it for Hotmail because we promised our mothers we wouldn't.
The first choice would mean an implicit admission that servers should be running FreeBSD and/or Solaris, but not NT or Linux, if they want real speed and performance. This does not sound like something Microsoft would say.
The second choice would mean, well, nothing. It's ridiculous.
Benchmarks, even when they are valid, should not be as important as the ability to achieve real-world results. Maybe it's just me, but I always thought that computers in general were supposed to be used for getting something accomplished as efficiently as possible.
And don't tell me that it was user error which caused Hotmail not to be able to run well on NT, which would mean that MS can't figure out how to use its own flagship OS. That would be pathetic.
Hotmail has been known to run Solaris; they tried NT - it didn't work out.
-Stu
'nuff said :)
-Stu
well i don't _lay it out_ in vi, of course! :)
i can type my thoughts more quickly without distractions... table of contents/paragraph spacing/fonts are all useless when you're trying to hammer out ideas.
-Stu
If this is supposed to be a comparison between IIS and Apache, then fine. But if its trying to offer a comparison between Linux and NT, its useless.
I suggest Mindcraft re-run the tests using a common webserver. Most web benchmarks I've seen use the Zeus webserver.
Here are my reccomendations for the boxes:
Wow, what do you know, NT server won't run on boxes 1 or 6, and on 2, 3 and 8 its gonna choke.
Okay, maybe I went a little overboard, but I'm sure you all get the picture. Dont' let this FUD distract you from the Ultimate goal of Linux... "World domination... FAST!"
Clinton made me a Republican. Bush made me a Libertarian. Trump is making me question reality.
Set OPTIM = "-04 -m486" before compiling
on 4 x 400 MHz Pentium II Xeon
Samba 2.0.1 Configuration
wide links = no
That creates a bottleneck in Samba performance, see here
the following processes were running ... (kswapd), /sbin/kerneld,syslogd,
not sure if that means something, but why they run kerneld with 2.2 kernel ?
On NT side:
Tcpip\Parameters\Tcpwindowsize = 65535
that makes huge boost on network performance, but only on local network where packets don't get lost
Set Logging - "Next Log Time Period" = "When file size reaches 100 MB"
Logs on the F: drive (RAID) along with the WebBench data files So basically server does much less logging than Apache - and since it's many small requests, and since Apache writes logs on a non-RAID disk all together it'll be a big bottleneck
Anyone noted anything else wrong with this benchmark ?
From all my experience it looks like pure crap
P.S. Why they needed NFS ? inetd ?
Obama 2012: our incompetent asshole is slightly less of an incompetent asshole than the other incompetent asshole !
I haven't looked at the Samba code, but the approaches you suggest are vulnerable to race conditions:
Your suggestion of doing an lstat() or readlink() before following the file is also raceable; I just have to do step (4) in between when Samba lstats bob and when Samba chdirs to bob.
In most situations, these symlinks aren't a problem; if the server is just a Samba server, there's no way for me to create symlinks, and if I have shell access to the server, I don't need to trick Samba to read any file on the system (that I have access to, anyway; symlink tricks won't let me read files I don't have access to anyway).
Not having used webbench for more then a minute before giving up, I would like to know what all these requesting machines get from the server.
The same file? I administer a farm of webservers,
including some very big machines hosting 400 plus
websites per machine. The biggest problem we have
is in disk wait time as no two requests are for
the same file.
Some file system statistics would be nice.
Only problem we have is when daily logs fill
up the hard drive and apache crashes the machine
cause it can't write.
System Activity Report for linux?
I always thought that was Mark Twain...
Run a pencil-and-paper RPG campaign with your far-off friends: Gametable!
This has been gone over here time and time again...
That setting is a system-wide setting only. What is being discussed is a per-process setting, which is in limits.h - default 1024.
NB: Per process includes processes with child processes like Apache.
Matt. Want XML + Apache + Stylesheets? Get AxKit.
Unfortunately the per-process max applies to all child processes too, so Apache would run into that limit when spawning lots of children. We had it happen to Postgres and setting the system max via /proc didn't help. Postrgres uses a similar process spawning model.
Matt. Want XML + Apache + Stylesheets? Get AxKit.
Who needs a web server like this anyway?
I just did some simple calculations:
At their maximum hit rate of 37,576,067 bytes/sec, that's 36,695 K/sec. At an average of 50K per page (that's a recommended amount for page size given download speed etc. It also happens to be about average) they're serving up 734 web pages every second (that's full pages including gifs, not hits).
In total that's 63 million hits per day. There's not a web site on the planet that has to handle that capacity from a single machine.
The point is that web server performance on static pages is bullshit. For some real performance figures try comparing ASP vs mod_perl and get the shock of your life.
Matt. Want XML + Apache + Stylesheets? Get AxKit.
I thought Twain said that. Or maybe they were both quoting a third party?
:-)
Daniel
"If you pick up a starving dog and give him food, he will not bite you. This is the principal difference between a dog and a man." -- this is Twain, I'm sure. Fortune sayeth so.
Hurry up and jump on the individualist bandwagon!
You're wrong. There is no version of Windows 98 that comes with support for the 3C905B out of the box. It supports the 3C905, but not revision B. I didn't believe it either until I saw it happen last weekend. Crazy.
--Defiler.
We started the tests using standard Red Hat Linux 5.2 but had to update it because it does not support hardware RAID controllers and SMP at the same time,
Ok, so they say that if you're using a RAID controller with Linux, you can't use an SMP kernel, and by the same token, you can't use a RAID controller if you're using an SMP kernel. For some reason, my bogometre is sensing an unusually high amount of bogons in that statement. Would anyone care to verify this?
^D
In Soviet Russia, Jesus asks: "What Would You Do?"
Except linux CANT use more than 960MB of memory without patches......
Unfortunately, the sloppy thing is that those items HAVE to be tuned. Linux needs to get rid of things like required patches for filehandles and the like, and support >1GB ram properly, before it begins to get good reviews.
I know everyone will whine and scream, but Linux on Intel just does not make a good enterprise server, it has too many serious shortcomings. And Linux on Intel is what will be tested, not FreeBSD or Linux on Alpha, because Linux on Intel is what all the linux advocates have pushed.
The problem is that Linux needs kernel patches for more filehandles, >960MB memory, etc. etc. It's very sad IMO.... a sysadmin should NOT be expected to have to apply patches.... menuconfig options maybe, but not patches.....
People complain about tests like this and DH Brown, but really only somewhat out-of-box solutions should be tested.
My understanding regarding multiple NICs in one box is that you need to specify how many NICs you have in your lilo.conf if you're using ISA-based cards. The PCI-based drivers automatically detect all of the NICs on your bus and you don't need to pass any special parameters to the kernel. I don't know if this holds true for all of the Linux network drivers, but I do know that this is the case if you're using the ne2000 drivers.
keep acting shocked and move slowly towards the cake.
Linux is getting MS so worried that they sponsor tests to discredit Linux. This is the beginning of a marketing campaign against Linux--Microsoft is going to War.
Free Software: the software by the people, of the people and for the people. Develop! Share! Enhance! Enjoy!
According to Apache, they stopped tuning for performance when a low-end machine could saturate a T1. The Linux box in the test peaked at 27Mbit/second. That is, the Linux box could saturate 10base-T, let alone anything reasonably between a web server and the clients. If you've got multiple T1s coming in, you may as well have multiple machines at the end.
For file service, the Linux box put out 114Mbit/second. That's more than 100base-T, and a significant chunk of a gigabit. If you've got this sort of thing on your network, file server performance is the least of your worries. Unless you intend to have a separate network for file service, like in their tests, this isn't going to be a big deal.
In any case, haven't we just seen that NT beats Linux on Intel-based high-end servers, but sucks compared to equivalent UNIX servers? Really, Linux isn't ready yet for the high-end, and this hasn't yet become all that big a deal.
"We work with you to define the goals you want to achieve via testing."
As their main web page states, they define the goals before they test. The only goal was to say NT runs faster than Linux. I've never heard of this company before. I now know why.
Somebody at Mindcraft must have had a really bad hangover. It seems like they have mistaken Linux for NT and vice versa. Just swap all the values in the page and it should be correct.
War is one of the most horrible things a human can be exposed to. And one of the worlds largest industries.
'Nuff said.
No sig. Go away.
Here's to hoping your post gets moderated up.
So they've intentionally crippled Samba (it boggles the mind to think of what they must have done to Apache) and lied about asking for help. I think it's safe to assume that this study is absolute rubbish.
-m486 is pretty standard. I would be very surprised to see a performance gain of more than about 1% between -m686 or similar and -m486.
wide links=no has been explained on other threads. It certainly slowed down performance by an unreal amount. This is a paranoia security measure that apparently some admins would use. I don't know that it's fair to assume that they put this in solely to skew the results.
As for kerneld, inetd, NFS, etc.: all right it's unnecessary, but will use under 1MB of RAM and under 0.1% CPU most likely. I don't see this as an issue.
My best guesses for the apalling results are something like this:
- the wide links=no thing. NT doesn't have to worry about symlinks. I think this is unnecessary in pretty darn well every case. This could either be intentional malicious intent for the pro-NT side, or inexperience/a mistake. Either way, it would be nice to show some number with this turned on.
- pure speculation here, but they may have set up Apache to do real-time hostname lookups. This is an absolute no-no for any serious server. Again, possibly inexperience or a mistake.
- the >512MB RAM Linux bug. I've heard horror stories, and I've read people with no problem at all. Also, I believe this was a problem with PIIs only, and they were using Xeons in this report. Who knows.
Anyway it appears to be a bad combination of very silly yet somewhat understandable (for newbies) software misconfigurations, and some bad choices in hardware. Which brings me to another point: quad Xeon for a server? My K6-166 could handle a few thousand hits a second I'm fairly certain. Adding more processors will only slow things down when you're dealing with file serving.
It's hard to say whether or not they did this intentionally. It's fairly obvious that they didn't know exactly what was going on with Linux. I'd say Microsoft has a list of hardware that they know works well, and when these people asked for sponsorship, some Microsoft people said "OK here's some hardware that we know works. BTW doing this and this and this might help out your performance". Not to say that Microsoft went out of its way to hurt Linux, but they probably know what works best on their own systems.
You are right. A conspiracy needs some sort of hidden plan/agenda. This is simply marketing.
-- Are you an EFF member yet?
Most of us don't have access to non production servers of the spec required in order to run benchmarks.
VA Research and Red Hat or Caldera do. It's in their business interest to run a set of their own test (or sponsor some).
Deleted
Check LinuxHQ and look either in the patches or the mail archives.
Wil
--
Internet Meta-Resources:
Wil
wiki
I tested file transfers mostly, and I can tell you they are about equal, with a very slight edge going to the linux box. With file transfers, the bottleneck is ususally the ethernet card.
I even tested the DES client on each box, which I figured should be nearly 100% CPU dependant. Still, a slight edge went to the Linux client, which suprised me.
(Sorry, I don't have the numbers anymore...)
--
"Reactionaries must be deprived of the right to voice their opinions; only the people have that right." - Mao
Mindcraft, Inc. conducted the performance tests described in this report between March 10 and March 13, 1999. Microsoft Corporation sponsored the testing reported herein.
The test figures do NOT reflect real-world performance behavior of NT vs. Linux. NT can't cope with the loads they're claiming- we've seen NT boxen just like the Dell supposedly used collapse under the load whereas a Linux box keeps on chugging.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
Check the daily updates section- it's there.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
And what's more, you're right. Wonder what it'd be like with a PROPERLY tuned test machine. They never once got to comparing apples to apples- and it seems they are guilty of past sins of this nature.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
Cut straight from their "White Paper"...
RAM: 4 GB 100 MHz SDRAM ECC
For those who think that the memory was a 1 to 1 comparison, you're mistaken. Whatever they told NT to do with the memory they did NOT tell NT to not use the 4Gb, it told it to use it differently. (Which, between the disparity of memory used and that, I suspect that this is where some of NT's supposedly stellar performance came from...)
Simply put, 4Gb != 1Gb. They should have stripped down the box to do an apples to apples comparison. That right there just invalidated their little benchmark.
Second, as others have pointed out, I suspect that Apache's problem was not threading, or Linux itself per-se. It was that they did nothing to really tune Linux on that box- the Apache test slammed squarely into a preconfigured limit (why it's there, I don't know, but it is, just the same...) for the number of open handles for files. That's why it seemed to peter out like it did.
There's just so damn many things that are sloppily done and dead wrong on the Linux side of things that I don't even really know where to begin taking this apart and proving it all the lies that it is.
I am not merely a "consumer" or a "taxpayer". I am a Citizen of the State of Texas
From my personal experience, I can not believe that NT should beat Linux in performance on such a massive scale as the thing we're seeing here.
But the fact that Linux is faster than NT does not imply, that it is the operating system/kernel of choice by IT professionals. We use GNU/Linux, because it works great, it's fast, and we won't have to spend money on licences every time we need another box running. But that is not the way an IT professional with a near-million-dollar budget thinks. If Linux can't ``utilize'' (the word we interpret as ``hog'' or ``crawl on'') a huge machine, then Linux is just not the system of choice, for running the really expensive/important server jobs.
Imagine running a million-dollar operation on 1000 dollar hardware. It's possible, and a lot of you reading this, probably do so allready.
But it is just not the way many IT pros think.
We will keep improving Linux and the GNU system, because we need it and love it. But there will always be people with too much money and too little brains, that will interpret the fact that the system also runs on small machines, as a clear sign that it is a system for small machines only.
NT is often considered a high-end server OS, and I believe it is because it only can run on high-end server hardware.
Those people, the ones with small minds and even smaller ideas, will eat performance comparisons like the one we've seen here. They will actually believe the results without even the slightest sceptism.
And the rest of us will keep running the world on the OS that has proven it's worth, not in the press, but in the field.
then of course the moderators won't ding it. You need to show the moderators that it's not worth seeing by demonstrating it's not worth a reply.
-----
One of the characters recites a quote, and two others argue about whether it was Mark Twain who said it or Samuel Clemens.
Pogo asks the character who made the quote, and he said, "You heard me. I heard me. I said it."
-----
HTTP Error 403
:-)
403.9 Access Forbidden: Too many users are connected
This is what I got when I tried to connect to your site, a NT box apparently running ASP.
So, is that what you mean when you talk about the quality of M$'s latest products?
Oh, what the hell, let me congratulate you for one of the best trolls I've ever read on Slashdot. Well done. Really.
This might sound as a mere flame or a personal attack, but from a strictly rational point of view, I find hard to take seriously anyone who praises M$ for vaporware ("Linux 4.6 will have though recognition devices!!! Tell me what M$ product can be *that* intuitive!!!!")
I looked up the list price on the Dell site. The closest configuration to the test machine was > 36,000 USD! I'm sure if you gave that much dosh to VA Research, they could knock together a demon.
I could even get a 2xUSparcII Enterprise Server 250 from SUN, with 2GB of mem (should be plenty,) and 45.5GB of RAID storage. And still have ~4000 USD to spare. So much for low cost Wintel!!
Why is it that when a website burdened with old hardware and an underperforming operating system posts news about a site running on NT, that the NT site can't handle the load?
Are they running that close to their peak all the time?
When their site is serving faster than one page in half-an-hour, could somebody tell me what they are running?
They aren't giving me enough throughput to fill up a 1200 baud modem here.
I'm not an expert on this, but I would have liked to have seen the test results on more modest hardware, such as a single processor Pentium II machine. (Not Xeon)
Everyone that I have ever talked to in the community say the SMP support in linux isnt crash hot yet.
More than something screwy. You can't use more than 100 threads on IIS 4. It uses Microsoft's Transaction Server thread resource pooling to do thread management, and MTS is internally limited to never handing out more than 100 threads. So basically at least some of it is incorrect.
"The following processes were running immediately before the NetBench and WebBench tests: init, (kflushd), (kpiod), (kswapd), /sbin/kerneld, syslogd, klogd, crond, inetd, bash, /sbin/mingetty [on tty2, tty3, tty4, tty5, and tty6], update (bdflush), and portmap"
do anyone see httpd there? so maybe they setup apache as inetd servicehany
yes, some tips would be nice. i'm going to setup my first web server for customer which will be used in real Internet (for now, i just made some LAN servers for 10 clients :) .
hany
According to ancient tradition, shouldn't that be called "Whebstone"?
-- Arm yourself when the Frog God smiles.
I agree completely.
It appears that the MSoft folks have discovered a very large hole in Linux performance, and that is that SMP is still not mature, or perhaps not sufficiently performance tuned yet.
As we have seen many other benchmark on single-CPU machines that contradict the results shown in this study, the only guilty party left to point to is the SMP implementation and tuning.
Of course the publishers of the study could be either incompetent or liars, but I don't think so.
I think they have discovered a large flaw, and are exploiting it in a very deliberate fashion.
Maurice W. Hilarius Voice: (778) 347-9907
"For Linux, we used Samba 2.0.3 as the SMB file server and Apache 1.3.4 as the Web server."
page 16:
"Mindcraft certifies that the results reported accurately represent the file-server performance of Microsoft Windows NT Server 4.0 and Red Hat Linux 5.2 upgraded to the Linux 2.2.2 kernel with Samba 2.0.1 running on a Dell PowerEdge 6300/400 as measured by NetBench 5.01."
Hmmm...
This report is rather hard to believe. Especially since all the tests done by zdnet and pc world shows linux to royally stomp NT in these two catagories.
I for one think varesearch or penguin computing should refute this report on their comparable hardware as SOON as possible as to avoid too many mis-informed people spreading this nonsense around.
I don't think microsoft honestly expects this report to hold up... I think its more of a attempt to cause reasonable doubt in the minds of would be linux users.
just my two cents.
-- You can be a geeklord too
Care to show your result on the SMB test you mentioned?
Thanks
-- You can be a geeklord too
echo 8192 > /proc/sys/fs/file-max
enjoy.
-- You can be a geeklord too
According to their report: httpd, smb & nmb were all running from inetd.
/sbin/kerneld, syslogd, klogd, crond, inetd, bash, /sbin/mingetty [on tty2, tty3, tty4, tty5, and
:)
There is no other possible explimentation if what they state is true.
The following processes were running immediately before the NetBench and WebBench tests: init, (kflushd),(kpiod), (kswapd),
tty6], update (bdflush), and portmap
What I can't figure out is what Redhat installation option sets up these services in inetd.conf? Very odd
-- You can be a geeklord too
echo 8192 > /proc/sys/fs/file-max
kernel patches my ass..
-- You can be a geeklord too
How many servers does it need to whump /.?
Name: msnbc.com
Addresses: 207.46.150.213, 207.46.148.249, 207.46.148.250, 207.46.148.251
207.46.150.201, 207.46.150.205, 207.46.150.209
Aliases: www.msnbc.com
Name: slashdot.org
Address: 206.170.14.75
Slashdot: Where nerds gather to pool their ignorance
Consider the odds... peaking at 1000, what a nice even number... I bet you one of those systems they used to benchmark, that there was some conf file that had a limit set. Perhaps after the 7th martini, the poor mislead man running the tests forgot a zero.
Earn cash in your spare time! Blackmail your friends!
If you look at the certification section it is "mindcraft" admits that MS paid them to do this benchmark. Frankly, I can't see how that could possibly make an objective benchmark.
If you look at the other whitepapers that this company has done it is very evident that they are highly biased towards NT.
Just look at the SMP Ultra Sparc machine getting
beat 4x over by some NT PC.
Over at linuxtoday.com one of the Samba team members gives other information
emad
Famous Last Words:
The only way to work this out (apparently) is if representatives of "both sides" came together and defined parameters of a test. They should specify things such as what hardware the server should run, what services should be running (and more stuff that you sysadmins know about for a living). The group should also lay down specifications on how this should be measured. Then, these specifications should be publically reviewed, and revised if necessary.
Then there should be a test for extremists on both sides to tune the OS of their choice to achieve the best possible performance on the specified platform.
If a vendor (say, Dell, Compaq, IBM, or one of the other companies that now presumably deliver both OS's) could put up a number of similar boxes and have a "tweaking contest" on a software convention or whatever, that would be even better. However, I cannot believe for a second that Microsoft would have the balls to allow such a shoot-out...
Yes, and they didn't even submit those ZD benchmakrs to ZD ie. they're not even certified!
If you want a secure system, wouldn't you be avoiding NT like the plague anyway? :)
Seriously, what are the ramifications of widelinks=yes? What is the security threat posed by this? Just curious...
~luge
IAAL,BIANLY
...I am dependent upon AltaVista!
-- LaTeX, The Best There Is
Excuse my not agreeing to your point of view, but I think that a sysadmin who _has_ to tune his server and an Operating System which lets him do it is the best arrow to shoot any deer. In a trade marked by continuous and sustained change, there is no way around. Period. Out-of-the-box solutions should be tested when you are looking for ease-of-setup/use results, but performance? No way!
--Moo
And Hitler had the VW beetle designed.
Seriously, though, there _is_ a point there about the inconvenience/immaturity/stupidity of Microsoft-bashing for the sake of itself: if all the energy there's been put into writing the 4million+ posts regarding this article (and the 2billion+ times somebody has read then) had been used to confirm the benchmarks, point out the wrong ones and correct linux/apache/samba wherever necesary and finally to publish new results, we'd be far better off.
--Moo
"I think people tend to make far too many biased statements when arguing either for or against Microsoft products."
You show yourself to be a good example of this phenomenon.
"Linux supporters invariably claim that Linux is faster and more stable, while Microsoft devotees tend to rely on the easier setup and configuration of Microsoft's operating systems and software."
Perhaps from your perspective. What about stability and the ability to fine tune and configure the system? Also, I think you are NOT talking about setup, but about installation.
"This test just goes to show that every operating system has good parts and bad parts."
No, this test just goes to show that Microsoft can pay for whatever results it wants.
"While Microsoft may be the current lead in speed and usability, Linux is still a little bit ahead in stability (notice I said "a little bit")."
Yeah, but you're wrong. Linux is much further ahead in stability, and given a fair test shows itself to be more usable and configurable (notice I said "configurable" not easy to configure).
"I think the main reasons for this are resources and one of the fundamental rules of capitalism."
Fine, let's hear your theory on how the world works.
"Microsoft has far more resources than Linux hackers could ever hope to have."
Wrong. Linux has far more developers that Microsoft has. Of course, they range from full time kernel hackers to those that do PR work on Slashdot and are innumerable, making this a unfair comparision.
"While they may not have nearly as many developers working on the product, they do have a relatively closely-organized team all communicating within the group and heading for specific goals."
Really? Are you sure?
"They have entire company departments devoted to testing, user interface design, code beautification, etc."
If this was true, so what? Creating a department isn't the same as actually achieving anything.
"Linux, on the other hand, doesn't have quite as organized a structure."
Hey, you got one sentence right! :-)
"This isn't to say that the development of Linux is worse or less efficient, it's just disadvantaged in some areas, by no fault of its own."
Wrong, that sounds like a fault to me, however you fail to describe any such fault. Furthermore, you fail to notice any positive attribute of Linux development.
"As for that major capitalistic principle, it gives Microsoft both an advantage and a disadvantage."
What principle would that be? Greed is good?
"Microsoft's goal is to make money."
I disagree. Microsoft's goal is more likely to control everything they can. The fact that they make money is just the result of that control.
"In order to attain this goal, they must have a superior product in order to get people to buy it (let's leave the monopoly crud out of this for a moment)."
Perhaps, but most likely they can better attain their goal by control.
"Therefore, Microsoft strives to make the best, fastest, and most usable product out there."
I don't think so. Being the best, fastest and most usable is impossible, since so many different aspects of software contribute to its quality, speed and usability. Microsoft is not suited for all tasks equally. I think Microsoft is best at being a desktop for users.
"And when they fail, they generally try to fix the problem as soon as possible via service packs, hot fixes, etc."
Yeah, sure. I think "as soon as possible" can have a lot of different meanings. It could mean:
"On the other hand, this is also a disadvantage because rather than implementing features that are needed by a few customers, they implement only features that are needed by a majority of customers."
Oh, is that why the system sucks so much? Why don't they make it more configurable and usable then?
"This is where Linux has an advantage, obviously, due to its open-source model."
Well, not so obviously but yes, I agree with the advantage.
"I won't get into it much further than this, but I think we should all at least give Microsoft the credit they deserve."
Thanks, I hate replying to people like you. By the way, Microsoft deserves no credit.
"For the most part, they do try to have a good quality product, and their latest work (e.g., SQL Server 7.0, Office 2000, NT 5.0/Win2000, IIS 5.0) has been of exceptional quality (at least, from my point of view)."
Yeah, they try to gain control of all the market's they enter by whatever means they can. Is Office 2000/NT 5 really that good? It hasn't been released yet and Office 97/NT 4 was supposed to be pretty good too. What didn't they just update their previous good product instead of forcing everybody to upgrade? WHY?
Why not view this as a challenge? I'd be willing to bet that Linux could catch up with and even surpass NT's benchmarks within a year or less.
WTF? Why would you ever change your mind? Doesn't Microsoft have an undeniable advantage?
And yes, I expect flames galore for saying good stuff about Microsoft...
And deserve them you do.
For all who are interested, here is another test done my Mindcraft. It shows NT being faster than Netware. It's a similar format, but some differences are apparent.
e svr.html
For instance, in this survey, they talk about price/performance, which isn't mentioned in the Linux survey (for obivous reasons).
http://www.mindcraft.com/whitepapers/nts4nw5fil
Citizens Against Plate Tectonics
Microsoft commissioned this test. That means that Mindcraft got their best tech support. This is normally a paid expense. You can see some pretty heavy registry hacks in their setup. You bet this kind of support would cost a bundle.
It appears as if Mindcraft spect $40 on a Redhat CD, and then expected Redhat to help them setup this complex system.
The postings didn't mention who they were or why they were doing this. Nor did it show any real information, other than the basic machine specs and the fact that it gets slow.
So, Mindcraft gets the top of the line, extra-expensive support from MS for free, but doesn't pay for the support from a Linux distributor. They then send in a fluffy posting pleading for help without providing any useful information. After all this, they complain that Linux doesn't have support.
If they bought their machine from VA Research or Penguin Computing or someone, or hired a contractor to help them set it up, you bet they would have gotten the answers they needed.
Citizens Against Plate Tectonics
Chilli
-=- Just a random lambda hacker
I agree that the article on Samba and NT interaction was good, but then, it was from one of our own, after all.
But did you read the hopelessly clueless little screed at the back from "Auntie M. Cee Pee"? Same trash as always - you can't use Linux if you don't know a bunch of arcane commands and a command line. Same old tired FUD at its worst.
What I wonder, for people who think that Un*x commands are arcane - have they ever tried to do any real performance tuning on NT? Talk about arcane...
Swampfox
Real Hacker (tm) Wanna-be
Deals
This test just goes to show that every operating system has good parts and bad parts. While Microsoft may be the current lead in speed and usability, Linux is still a little bit ahead in stability (notice I said "a little bit"). I think the main reasons for this are resources and one of the fundamental rules of capitalism.
Microsoft the current leader in speed and usability? Speed? Every other benchmark I've seen with Linux pitted against Microsoft has shown Linux coming up ahead.
I am a consultant in my spare time, and I've set up a company with a dual processor Dell Poweredge server with 512MB ram that has been serving over 100 clients via Samba with no problems, and the server has enough power left over to do E-mail serving, IMAP serving, POP3 serving, web serving, and a myriad of other smaller tasks, and the load doesn't even hit 1 during the busier times. If you take a look at this MRTG graph you'll see that the server sends out a hell of a lot of data, and it's been a great primary domain controller for this place.
Linux has been phenomenal, except for a couple of small problems with the system running out of file handles. How can I tune the system a bit to get rid of these pesky little problems?
Does anyone have any suggestions where I could go?
Ben
I just wanted to mention that I've recently set up a Dell Poweredge server with PERC RAID controllers, 512MB RAM, Dual processor Pentium II 450s, and Linux 2.2.5.
This machine serves over 100 clients, and it functions as a primary domain controller running Samba 2.0.3. It has worked phenomenally for the all NT network my client has, and it also serves Email, IMAP, POP3, and is a web server for all the users here. It also does a myriad of other tasks, and the load never even hits 1. And this thing is a less powerful machine than they tested, but it can serve over 100 clients with ease.
I don't know where they cooked up the figures they have, but this server gets plenty of use, and it's never buckled or given me any problems setting it up.
By the way, does anyone know where I could go to find out how to increase the maximum number of files, and/or to further tune this machine, because I've had a couple of small problems with running out of file handles for the whole system. Anyone have any suggestions for a site I could go to?
Yeah, this study is sponsored by Microsoft, if you read the fine print:
Mindcraft Certification
Mindcraft, Inc. conducted the performance tests described in this report between March 10 and March 13, 1999. Microsoft Corporation sponsored the testing reported herein.
Looks like you can buy anything you want with enough money. It doesn't make it a true indication of a real-world situation.
I think that there's enough evidence to the contrary already out there, and this will only serve to discredit Mindcraft.
According to a posting on Linux Today by Jeremy Alison of the Samba Team, it seems that the Mindcraft study crippled the Samba server in the tests:
:)
From Andrew Trigell (original Author of Samba):
They set "widelinks = no" now I wonder why they did that
In case you haven't guessed, that will lower the performance enormously. It adds 3 chdir() calls and 3 getwd() calls to every filename lookup. That will especially hurt on a SMP system.
I think it is important that the Linux community contests these results, as they certainely seem skewed, but I hope we can do it calmly and with dignity.
With the amount of equipment involved, I believe it would take VA Research or a company of that ilk to try a similar test.
(Could we do a test on a less expensive set of equipment?)
If the results of the original report are not reproducible, then what they did is bad science. I think that trying to reproduce the results, but with people who know how to optimize the Linux set-up (and be fair and optimize the Windows set-up too) would do much more for how Linux is preceived than by us doing a lot of name calling and questioning the motives of Mindcraft.
The point is, if everybody else who does the test gets completely different results, that will be all we ever need to say. (publish the results of course)
Let's respond in a way befitting the wonderful operating system that Linux is.
I think it is important that the Linux community contests these results, as they certainely seem skewed, but I hope we can do it calmly and with dignity.
With the amount of equipment involved, I believe it would take VA Research or a company of that ilk to try a similar test.
(Could we do a test on a less expensive set of equipment?)
If the results of the original report are not reproducible, then what they did is bad science. I think that trying to reproduce the results, but with people who know how to optimize the Linux set-up (and be fair and optimize the Windows set-up too) would do much more for how Linux is preceived than by us doing a lot of name calling and questioning the motives of Mindcraft.
The point is, if everybody else who does the test gets completely different results, that will be all we ever need to say.
Let's respond in a way befitting the wonderful operating system that Linux is.
http://linuxtoday.comf tcheck.html
also provides a link to an article about a similar incident Novell had with Mindcraft.
http://linuxtoday.com/stories/4937.html
http://www.novell.com/advantage/nw5/nw5-mindcra
The one on the novell website is epecially informative. It is the exact same situation in which results published by Ziff-Davis show NT at a disadvantage, but when Mindcraft does the test, NT comes out ahead!
I haven't been able to get at the whitepaper yet due to it being slashdotted, but I was able to read the press release. It looked fishy to me that they were only comparing peak throughput. Does the whitepaper say how they defined peak throughput? It could be just a statistical aberration if they weren't careful. In any case, I hope they were honest enough to use sustained peak throughput.
I have personally benchmarked Linux and NT machines on our UCLA campus network. My friend has put his Linux P133, 40MB ram IDE hard drive print server up against our very own www.resnet.ucla.edu machine that is a dual P2 400 256MB ram, running NT and dual Seagate Barracudas in Raid 1 configuration. These two machines were on the same physical hub on this network for this test. The NT machine could beat the Linux box up to about 20 simultaneous http connections(only by about 20% though), after that the Linux box smoked the crap out of that NT machine. The NT machine really started to take a dump after about 30 clients, and would start dropping connections. And speaking about kicking the crap out of something..... don't even ask what happened when I put my P2 266 box up against www.resnet.ucla.edu. The results were one of the biggest jokes I had ever seen.
:)
I don't believe this test data at all. I have personally seen NT boxes stop responding after Slashdot took them down. I can't even believe that the NT machine even accepted that many simultaneous connections. Our NT http server just decided to stop responding to any of them. Sorry about my crappy spelling, but I am really pissed off at this FUD.
And if anyone wants to do any http benchmarking of their very own, the machine I'm sitting at right now is fair game.
My Linux box here P2 266 96MB ram, IDE drive
Try not to take this NT box down, I don't own it, I just linked to it
The NT box www.resnet.ucla.edu here dual P2 400 256MB ram, SCSI RAID 1
Time for 6th grade English again?
Did you know that when people post through DejaNews, the originating IP is included in the headers?
For our boy will@whistlingfish.net, any linux group, we get exactly 2 posts. One was made to comp.os.linux.setup, for crying out loud. I love c.o.l.setup, but don't you think c.ol.networking would have been more appropriate? But I digress.
The posting IP for will@whistlingfish.net is 131.107.3.71. nslookup 131.107.3.71 yields... drumroll, please! ...
Host name: tide71.microsoft.com
IP address: 131.107.3.71
Alias(es): None
Heh heh. Yeah, "we're thinking of going with Apache on Linux," you bet.
jerrod asks what OS was really running with this user_agent string:
X-Http-User-Agent: Mozilla/4.0 (compatible; MSIE 5.0b2; Windows NT 5.0; MS ITG IE5 internal).
Note the last word, "internal".
Now looky here. The originating IP of these posts was 131.107.3.71. Guess who that is.
nslookup 131.107.3.71.
Host name: tide71.microsoft.com
IP address: 131.107.3.71
Alias(es): None
The poster works at Microsoft, so he/she/it could very well be using a pre-production version of NT5 a.k.a. W2K.
I dunno what is wrong with your Apache, but my production Apaches always run with 30000, and I use mod_perl. No leaks have been observed.
-jwb
This is actually known as a problem with certain motherboards using the Intel chipset, which I've run across in my travels. Although the guys at Ars Technica (www.ars-technica.com) would probably know more, I do know that the best you can do on such boards is try to work around the problems with OS tuning, but the "real" fix is to get a better board ;)
Obviously if you know the workarounds better for NT than for Linux, you are going to get better performance on the former.
Nothing worth doing is worth doing today.
Looks like it.. Similar to other reports, but favouring NT instead
If you're going through hell, keep going. -- Winston Churchill
so your pitting NT against Linux. Does anyone else see the sense in this? They're pitting a FREE OS against an OS that has the force of GIANT, COLOSSAL software company behind it.
I mean, take as it is now, Linux is completely free, NT isn't. That's one of the biggest bitch to deal with. 'Sponsored' or not, I think this is still an unbelievable milestone. In a few years, the competition (NT) won't even be able to hold up to Linux.
NOTE:
The fact that Linux doesn't have as much to lose as NT does...
The fact that the Linux setup is MUCH MUCH cheaper than it's NT counterpart...
As far as I'm concerned, for now, these SPONSORED events can 'lie' or 'misconfigure' the system all they want....
I suggest we don't get our titties twisted in knot... give it a while, at this rate, Linux will prevail
--
the belief of one is more powerful than the interests of a hundred..
Just to refute one point, I'm running Linux SMP right now, and I've worked in labs where Linux SMP has been used extensively, and as of 2.1.x/2.2.x, SMP support is definitely not weak.
--
Ian Peters
dont[root]:~# queso www.mindcraft.com
209.218.193.11:80 * Windoze 95/98/NT
Go figure.
---
--- http://foo.ca
Actually, IIS on NT handles more big sites than Linux does. MSNBC is the number one news site on the net, whomps slashdot by several orders of magnitude, and it runs on IIS.
Reality check!!!:
msnbc.com. 1H IN A 207.46.150.209msnbc.com. 1H IN A 207.46.150.213
msnbc.com. 1H IN A 207.46.148.249
msnbc.com. 1H IN A 207.46.148.250
msnbc.com. 1H IN A 207.46.148.251
msnbc.com. 1H IN A 207.46.150.201
msnbc.com. 1H IN A 207.46.150.205
slashdot.org. 1H IN A 206.170.14.75
Have you seen what M$ recommends for a large site? They suggest using a cluster of quad-proc systems to support what a properly set up Linux machine can do with one. The smallest recommended IIS installation costs upwards of $250,000, software not included.
Comparing a dual PII with 7 quad Xeons will not give you adequate information about the software running them, nor the load being generated. Slashdot generates quite a bit of load for every page, and handles it all locally. MSNBC uses something like 20 times the hardware to provide the web serving and ASP. The database is likely stored on another half-dozen machines we can't even see.
GStreamer - The only way to stream!
Well, I don't know about 4 NICs, but I'm running a 2.2.2 kernel with 2 network cards in a few machines, and in all the configurations, I have compiled the network drivers into the kernel. One machine has a 3C905b and a WD80x3, and the other has two 3C905b's. RedHat's configuration has nothing to do with it, once you move to a custom kernel build, and I never build anything as a module unless I'm only using it sparingly (like SCSI Tape and SCSI Generic).
I used to think printing on on Unix sucked. Then I figured it out. Printing on Unix *does* suck. Like a Kirby.
I actually used to work for a market research firm that was hired indirectly by Microsoft (via one of their hired PR firms but paid via a Microsoft check) to do a survey for a product launch. Having viewed the communications (and never being asked to sign an NDA), I should confirm that Microsoft (via their plausible deniability proxy) was the only company we dealt with that defined what questions should be on the survey based upon what answers they wanted to receive. Other companies may have questioned results that didn't fit their image of what their customers wanted, but Microsoft was the only one that wanted to tell us what the customers wanted prior to us even asking the customers.
IIS is horrible. All the configs are in weird places, it logs in weird places, asp pages run the client machines like mad...sometimes inetinfo just up and forgets that it's supposed to be serving pages. Oops! And that's only when it's limited to 200 connections....
bah. run linux. you need stuff from nt you don't want to move? smbmount is there for you, babe.
--Mandi walls@NOSPAM.juniata.edu
aka kerolin
It seemes interesting that Microsoft seemes to think the anti-trust trials are almost over, and it's back to it's old tactics.
Two things:
- Yes, there are plenty of MS-only shops, or shops that are "close enough" to be primarily worried about serving MS clients (NT/Win9x). I think it's fair to benchmark "Server X with NT/Win9x Clients."
- Second, it's already been shown that Linux/Samba is still faster than NT, even in an otherwise MS centric shop, mixed in with other NT servers.
I suspect that they've crippled the Linux server somehow (disabled SMP in the kernel or used a Really Old version), or carefully selected benchmarks that rely on features NT is better at. I've seen plenty of tests (done some myself) where two systems perform very differently depending on the nature of the benchmark.Your Servant, B. Baggins
Dude,
you have a superiorty complex. get over it.. slashdot is not real life.
Ofcourse I cant spell..
Cheers
--
According to the Ethernet-HOWTO you need to specficially tell the kernel you have 2 or more ethX devices if they exist.
--
I researched and found the same thing. I'll be using a higher-grade toilet paper this week after reading the report. I'm certain this is simply the ms press machine at work.
TO ANYONE POSTING IN THIS THREAD:
Please, we know microsoft does these things, let's PLEASE try to keep on topic here. Criques of the article only, not microsoft's *cough* media strategy.
--
>Linux 2.2.x has the same default window size. The memory limit is hard coded in Linux, unless you apply some patches.
I'd beg to differ here - passing mem=000M via lilo will cause linux to address that memory - provided the system has that much addressible space.
--
I sure hope this benchmark doesn't stop
Linux from coming out and getting popular.
-fb Everything not expressly forbidden is now mandatory.
MINDCRAFT, INC. SHALL NOT BE LIABLE FOR ERRORS OR OMISSIONS CONTAINED HEREIN, NOR FOR
INCIDENTAL OR CONSEQUENTIAL DAMAGES RESULTING FROM THE FURNISHING, PERFORMANCE, OR USE OF
THIS MATERIAL.
Solaris I believe, after giving NT a try. I think NT is in there somewhere, however, because hotmail is multi-layered.
See that "Preview" button?
You get worst than that if you go to some
:"designed
of Microsoft's sites. This afternoon I was
looking for some information on one of Microsoft's
site. I used the search command, a tragic mistake,
it sent converted Netscape into some kind
of Zombie state. Netscape was still there but was
incapacitated. I would close it and come back
and it was still in the same shape.
I gave the winblows salute to get the kill menu
and saw netscape was in the task list even though
I had closed it. I kill the application and
reloaded netscape which was now running correctly,
until I went back to the Microsoft's site again.
I had noticed this time that it said
with J++"
What is tha garbage software by the way? I
thought that it was supposed to pull off the
market.
But where is this proof shown? This report has charts and graphs much like what a manager would want to see. If you go to http://www.unix-vs-nt.org you will notice you have to do a lot of searching to find comparisions between nt and linux. Most of these comparisons are out of date. In other words a manager might rely on this report especially when cnet is reporting it and there is no counter to it.
/. it? I have never used smp in my life (not enough money) and would never be able to do this type of comparison but there must be some way to do this and prove one way or the other which one is real. And I bet that there are a lot of people here that know about bench marking and which ones are the best. Is this at all possible or am I living in a dream world?
Is there anyway what so ever to put this to the test and do a real world test. Pit a nt server against linux server and
P.S. I can't spel
Having Linus putting his own quad to the test would be great. But would he be willing to lend his machine to this and would we be able to find the same configured machine but with windows nt?
Plus I don't know him on any personal basis and would feel foolish just sending him email out of the blue. I was more or less hopeing that some company that needed a server would be able to donate research money and have some people test out what is the best to buy. (research == practical real life tests with various bench marks)
P.S. I can't spel
There was a post to the linux-kernel mail list, which said that nt was optimized where linux wasn't but further said that there might be problems. I love linux don't get me wrong but I went to the sites you recommended but I saw nothing that had tests done with a comparable machine. If someone is going to set up a server then a machine like above would be great. There has been a lot of posts on the linux-kernel lists about how to further develop the smp parts. I don't know if something happened or that the people testing didn't know how to set up the linux kernel correctly. All I am saying is that we should get some numbers that are on high end server systems since this is what a manager would be more interested in instead of a small system that would be more likely a work station.
P.S. I can't spel
Who says that they have no clue? I thought that they managed it very well to completely mis-tune the Linux system while not making it too obvious. Read this too and think about it. They claimed that they asked in newsgroups and mailing lists, but they did not mention in their report that they fine-tuned the file buffer, for example. I wonder how many manager-types "bought" this story. One can only hope that they'll install NT instead, pay far too much for the performance they'll get out of it, and never find out why other companies can operate better at a lower cost. :-)
"I love my job, but I hate talking to people like you" (Freddie Mercury)
"So Netscape dies trying to run a little [unauthorized program from an unauthorized, but highly suspicious, source] and you blame Microsoft's web server? [...] Run [one that allows any Joe Blow and his Herd of Minions access to your every registry setting]."
Mmm hmm.
I am not sure why everyone says M$ stuff is so easy to install. I can honestly say that after installing NT about a million times I went to install it on a laptop that was NT complient or whatever they want to call it (had the little 98/NT sticker on it) and I can tell you it took me about the same amount of time to isntall redhat 5.1 (and this was my first time installing linux period).
:) In regards to how everyone says nt sucks his point is that if you understand the os you can make it do anything. I agree 100% it is just funny how everyone assumes that NT is easier when that is not always the case.
Then you come to the issues of sound cards that are plug and play that are isa... what do you do cause lord knows they don't work right out of the box. You gotta go in there and (I forget the file) but you have to run the plug and play file for isa cards. Kinda silly and not very well documented although in most cases you can find this with the card manufacturer.
I am not saying I am a NT geek or anything but I used NT enough to know how to install it and for people to say it is easy to install makes me laugh esp since I installed 5.1 of redhat in less time then I installed NT after a few million times of installing it.
But then again like one of my co-workers would say (he can make NT do anything arg) "I can really make linux suck to you know."
--MD--
--MD--
(The following is a vague memory, use it for the pint and not historical accuracy, please) .The actual test was a two car comparison. . .
During the Cold War, a Moscow newspaper released the results of an International comparison of automobiles. The Russian car finished in a close second, while the American car almost finished dead last. Moreover, while the Russian car won outright a great number of categories, the American car finished dead last in many catogories. .
I am also somewhat unsure how fair the test would be if the people conducting the test were knowledgable enough to set keys in the Windows registry, but not enough to read documentation and upgrade a kernel without trashing a system.
'There are 3 types of lies, lies, damn lies, and statistics' -Samuel Clemens
Unless, of course, that ninny's on Bill's payroll -- but then he's just a liar.
DFL
Never send a human to do a machine's job.
Because these "benchmarks" are a tissue of lies that could not bear comparison to real world usage of NT vs. Linux. My P120 with 48MB of RAM running Linux is clearly faster than a PPro200 with 64MB RAM running NT. Actually, I think my Linux box is faster than NT on a PII-350 with 64MB of RAM, too -- but that's a tough call. Without question the redraws are handled better with Linux, though. There's no comparison.
DFL
Never send a human to do a machine's job.
Folks... Pat yourselves on the back. we have made it to stage 3 of the (ignore/laugh/fight) scenario. Microsoft is plenty scared now that the truth about them having no real development road map has finally leaked out. Paying for a study is an amusing and pathetic way to get attention.
Microsoft will be served papers soon by my company
to immeadiately cease and desist using the words Open Source in conjunction with their products.
I've contact Ralph Nader already and he is aware of this. Ed Muth said recently "...there are many definitions of Open source."
No Mr. Muth, there is only one. And it is a registered Service Mark. I suggest you conduct yourself accordingly.
Cheers,
Nick
Microsoft used to (and presumable still do) the same thing with Novell Netware. All these press releases come out showing that NT does something better than Netware in an "independent" test, but you see that Microsoft commissioned the test when you read the fine print. Novell came out with their "Reality Check" series of documents to rebutt these studies. Some of these Reality Check documents are really interesting. See the new one on their site documenting BorderManager vs MS Proxy .. pretty hilarious.
Jason
"FORMAT C:" - Kills bugs dead!
In fact, you don't even need to edit a few lines in the kernel; just apply the latest 2.0.37pre patch; large memory configuration (up to 2GB, I think) is part of that patch.
As for the other bit about the swap, I'd LOVE to see Windows (not) working without a swap/virtual memory...hehe...
Apache is NOT written for maximum speed. The Apache Group developers openly acknowledge this. If you really need maximum performance out of a box serving static web content (which, in practice, almost no one does) you should look at Zeus (http://www.zeus.co.uk/) or the open source thttpd (http://www.acme.com/software/thttpd/). Both of these will do MANY times better than Apache under a heavy load.
Also keep in mind that, as the Apache Group says at http://www.apache.org/docs/misc/perf-tuning.html, most people can easily fill all of their bandwidth running Apache on a low-end Pentium. You'll run out of bandwidth long before you run out of power.
As I read the configuration, the NIC configuration really jumped out at me. They explicitly set the NICs to 100Mbit, and set them to bind to one CPU each. Now I don't know much about the EEPro 10/100, but will it automatically use 100Mbit on Linux without explicitly using it? As well, if they didn't bind the NICs to each CPU in Linux, won't you get a bottleneck as all 4 cards fight for the CPUs?
/etc/conf.modules and kernel boot line.
I also noticed that the NT box was tuned by someone who is obviously very well versed in NT internals. The Linux box appears to be an out-of-the-box install, with the proper settings turned on. They didn't even include the
One more thing, notice the cache TTL and max number of open files for IIS. Both are set to very high settings, which will ensure that files will not be paged out during the runs.
Good point, but one might also wonder why we should bother responding at all. Are we supposed to respond to every bit of FUD that comes out of Redmond?
IMHO, the justice case has shot their credibility more than anything we could say about them.
I hope people realize this piece to be the FUD that it is... and if people don't realize it.. then they will just be that much further behind in the revolution that is linux.
This brings out another problem.. if someone does a similar test that shows f.ex. that Linux in fact is faster, it will without doubt be turned down if those who make the test somehow are related (directly or sponsored) to OpenSource or Linux.
vr
well thats the biggest bunch of b.s. i've ever read. first of all, apache can carry a heck of a lot heavier load than that on a p 133mhz with 96mb ram without any problem except for it being a little slow. sorry, i may be reading all these graphs wrong, but it seems to me that they make nt look better by testing more nt systems. of course thats gonna make the performance look sweet. i mean, ok. you take 20 nt computers and 5 linux computers. linux beats nt by 2x. that would still leave nt looking better since only 5 linux computers were tested. maybe i'm wrong, but i think they are trying to pull the wool over a lot of peoples eyes.
heh, guess they had to use such a massive ammount of hardware to test with since nt usually crashes. whats microsoft's slogan again???.....i think its, if nt server crashes for you, microsoft says, "Buy more RAM." hrm, bill gates, what a great guy&&@$&$#@%$^$#^$&$^$. anyways, windows sucks and umm, linux rocks. i feel the force is strong in this one. over-and-out, rage-dca.
Yep, and even Twain attributed it to Disraely.
They specifically listed the processes running.
X, XF86_WHATEVER, WindowMaker etc, wasn't on the list.
Besides, the problems I was having were in NT5 using IIS5 on a P166 w/64 megs of RAM...not the most optimal system configuration for such things.
That's a kick-ass web server configuration for Linux, FreeBSD, OpenBSD, BSD/OS, etc.
It only sucks with NT.
Hell, tests even show that Samba on NT beats regular NT networking.
A 6350, which is the rack-optimized 6300, and only 2 Xeons (1MB cache). It's going to run Oracle and a bunch of other stuff. We have run Oracle on NT for years (3.51 mostly, which we've found much more stable than 4.0), and don't want to subject ourselves to random crashes anymore.
/proc/sys/kernel/file-max, already mentioned. This is in the Apache FAQ.
Things that jump out at me:
echo 2048 >
The MegaRAID driver was at version 0.94 at the time of the test. Had they asked, Dell would have sent it to them. But really the MegaRAID is a poor choice; for less than the price Dell charges for a MegaRAID bundled with a server, you can mail-order a Mylex AcceleRAID ($1700-some and in stock at buy.com), which is faster and better supported under Linux. It is true that even today, with the 0.96 driver in the 2.2.5 kernel, you will find only version 0.92 on ami.com. This is an excellent reason not to buy from AMI.
Warning: Do not buy a PERC II card from Dell today. In typical Dell fashion, they are shipping
some Adaptec 789x thing that has no Linux driver without changing the part number that used to refer to a MegaRAID. This is one of many excellent reasons not to buy from Dell. Why do we continue to do it? Well, we've bought some VAResearch boxes, but they simply can't match Dell's ability to throw hardware at you quickly if something is wrong. It doesn't matter how clueless Dell is if you know what you want, and their cluelessness can be an advantage at times. Their sales droids are not always the brightest, but there are lots of them, they return your calls, and they will jump through hoops (if you explain how) to make you happy. Also, Dell's default server warranty is 3 years, first ON-SITE. That's a big factor if your power supply blows up at a bad time. I think Penguin is catching up, though. Maybe next time.
The article mentioned searching the web for the most recent kernels and stable module versions... I'm sure it would have run much more smoothly if they had compiled their own kernels instead of mix-and-match from several places.
Surely you can't be serious? With billion dollar Bill on thier side who has to worry? Sue all you want and we'll tie you up in courts for as long as they like.
These tests must be verified by others. Shame they used such pricey servers. I would love to try this myself.
I have serious problems with Mindcraft's method of aquiring there assistance. The 3 usenet half-hearted requests for help is really disgusting.
-=[ KMFMS & KMFMC ]=-
DrWatt
Uhh I sure hope you're just not thinking. Keep in mind, ftp.cdrom.com set its records using an Intel EtherExpress card.
The revolution will be mocked
Whilst we're still comparing raw speed of SMB systems are we forgetting other improvements. Systems such as CODA add further benefits which may outweigh a slight speed improvement. I know that CODA is being developed for NT but in terms of real business use perhaps considerations such as that ought to be taken.
Notice that the log file location is on the RAID drive (F:) for the NT/IIS configuration.
No mention is made of where the log files were on the apache test, which I assume to mean that they are in the default location (the same drive as the OS).
For the type of test they did (many small http requests), would the disk i/o of the log file become a bottleneck?
This is the introduction to the "general performance tips" section of the apache manual.
------------------------------
Apache Performance Notes
Author: Dean Gaudet
Introduction
Apache is a general webserver, which is designed to be correct first, and fast second. Even so, it's performance is quite satisfactory. Most sites have less than 10Mbits of outgoing bandwidth, which Apache can fill using only a low end Pentium-based webserver. In practice sites with more bandwidth require more than one machine to fill the bandwidth due to other constraints (such as CGI or database transaction overhead). For these reasons the development focus has been mostly on correctness and configurability.
Unfortunately many folks overlook these facts and cite raw performance numbers as if they are some indication of the quality of a web server product. There is a bare minimum performance that is acceptable, beyond that extra speed only caters to a much smaller segment of the market. But in order to avoid this hurdle to the acceptance of Apache in some markets, effort was put into Apache 1.3 to bring performance up to a point where the difference with other high-end webservers is minimal.
This is just like television, only you can see much further.
I don't know about SQL server 7.0, but SQL server 6.5 was one of the biggest nightmares in my entire computing career, surpassed only perhaps by the experience of 'upgrading' to IIS 4.0.
Okay, folks. So we have a bit of egg on our face for this one, because nobody (to my knowledge) has really stepped forward with large-server Linux benchmarks which demonstrate anything differently. It may be that Mindcraft royally screwed up, or it might be that Linux really is slower than NT for a certain set of benchmarks -- the truth is more likely a combination of these factors.
If Linux is going to be treated as a serious operating system by the majority of the IT community, it's going to have to step up to the plate and demonstrate scalability and performance which does rival NT server in this area. Most of our knowledge about Linux-vs-NT performance is somewhat anecdotal -- we haven't really "put our money where our mouth is" and shown objectively that Linux can outperform NT in these areas.
Rather than dismissing this study as FUD, I think we could learn a few valuable lessons from this study. We should seek to understand why the benchmark results weren't as great as we would have liked. We should fix any obvious bugs or misfeatures in Samba, Apache, and the Linux kernel which stood in the way of higher performance. And we should stive to improve the entire system to make it be a true NT rival.
We have a lot going for us. First of all, we can innovate at a much more rapid pace than Microsoft -- so hopefully within just a few short months (and I'm being pessimistic!) we could demonstrate a high-performance Linux file and Web server which kicks NT's butt all over the place.
Nobody said building a high-performance, scalable Internet server operating system was easy. Let's get to it!
Matt Welsh, mdw@metalab.unc.edu
I haven't seen this mentioned yet, but the list of processes running mentioned by Mindcraft looks like what you'd get from ps aux, and they don't mention either smbd or httpd. This could mean that they were running the servers out of inetd - if so, I'm really surprised that they got the results they did, because I would have expected it to choke well before it got as far as it did.
This doesn't make any freaking sense. This is what happens when you have someone who doesn't know what hardware to buy and how to configure it. I would love to know what the guy who configured did for a living.
(An NT tuner no doubt...)
Numbers? You want numbers? We've got numbers.
, 387506,00.html
6 321,00.html
0 22,00.html
Our Jan. NT v. Linux headliner:
http://www.zdnet.com/sr/stories/infopack/0,5483
New Samba on PacHiTech v. NT
http://www.zdnet.com/sr/stories/issue/0,4537,39
Linux v. Netware - The New King of the Hill: Linux
http://www.zdnet.com/sr/stories/news/0,4538,398
Steven, Senior Technology Editor, Sm@rt Reseller
I work in tech support...90% of computer users are morons.
There was a discussion about SMP and heavy I/O limitations in Linux 2.0.X a while ago on Slashdot, so yes it was a valid problem. This was corrected in 2.2.X.
For more info, search for "Ring 3 Scaling problems" I believe.
This may be true for a very light load, but all the other benchmarks I've seen have concluded that Linux beats NT only after ~32 clients - from here Linux quickly outpaces NT. The main concern with server performance is how well it can handle normal-high loads. In other words, if you have a very light load (32 clients) either will work fine. However, if a high load is anticipated, the *independent* studies that I've seen seem to sugest Linux/UNIX and Apache. Besides this, if you aren't in a closed system (ie a lab) there are too many other variables to really know if lag you see is actually happening on the Linux box (and not network congestion, routing differences, etc)
\forall code \in C, \frac{\Delta readability(code)}{\Delta t} < 0
Microsoft uses a little tree widget in some of their web pages. On IE, this renders nice and quick as an ActiveX control. Takes about an eon to load as a Java applet in Netscape. (Although, as far as I can tell Linux + Netscape + Java = lock every time.)
--
Business. Numbers. Money. People. Computer World.
Note this is the exact same FUD that Microsoft used to pull when it was battling Novell for the File+Print market.
There would consistantly be benchmarks on two processor systems, even though Novell wasn't even touching the second CPU (this was the P5 era, when 2 processors were very rare.)
On the other hand, I don't think anyone has any doubt that Linux would out perform NT on a x86, where x = 5, but the type of people that see these benchmarks generally don't care about that. Benchmarks on "typical" server hardware (say a Proliant 5000 with 2 Pentium Pros and 256MB ram) is much much more interesting.
--
Business. Numbers. Money. People. Computer World.
Actually, most of the performance tuning on the NT side is rather vanilla, but this is the one that struck me as being odd.
The Affinity tool allows you to bind certain processes to certain CPUs. I'm not quite sure why they'd do this with the network card, except perhaps to fix some deficency with the NT networking code when the cards are running full bore. (Which as someone pointed out, they were saturating normal ethernet.)
Anyone have any other ideas why this was done?
--
Business. Numbers. Money. People. Computer World.
Well assuming the PR audience for the bench is Microsoft shops (because Unix/Linux shops hardly would care), I think the test makes a valid point. Linux is no silver bullet for your high end NT/IIS woes.
In fact, it's probably correct that a tuned NT/IIS system out performs an untuned Linux system. If you currently have NT/IIS, it's probably is a hellva lot smarter to try to tune what you've got before junking it for something else that you would have to go back and figure out how to tune anyways.
(This of course doesn't address IIS's instablity, which is the number one bane of IIS shops, not throughput.)
--
Business. Numbers. Money. People. Computer World.
Not a very helpful reply, eh?
Actually, I suspect the real high end box Apache tuners are out making big consulting dollars and not reading usenet. Think I'll go post a question about processor affinity and network cards on comp.os.ms-windows.nt.networking, just to see what turns up.
--
Business. Numbers. Money. People. Computer World.
Even if this were true, they were using a microsoft Client to talk to a Microsoft server, and using samba instead of something like NFS. It stands to reason that a Microsoft file server doing Microsoft file serving might perform better than a Linux box doing Microsoft file serving. I wonder how the tests would have turned out if all of the clients were Linux boxes, and a non-MS oriented file server setup were used.
-lx
If their NT is such a great server, lets just all go take a peek at their site...maybe hitting reload 50 times or so...
Mindcraft
-lx
You do realize that having Linus do this test would look roughly the same as Microsoft sponsoring the Mindcraft study. Truly independent entities who aren't sponsored by any special interest (i.e. Microsoft or Linux backers) would seem credible to me.
Thats exactly it. This Mindcrap group can say whatever they want. If their tests aren't independently verified, their words ring hollow.
Those in the medical community know that when a medical study's results are published, more testing must be done by others to corroborate results before it is accepted as true. Yet the media will happily latch on to the results and preach them as the gospel truth. It's a shame though, because a lot of managers in companies put a lot of faith in those media sound-bites.
Apparently they were smart enough to de-tune Samba, but not smart enough to tweak the rest of the Linux box. I would suspect that Microsoft said "Hey, Mindcraft, see what it takes to make NT whoop Linux! If it works out, we'll publish it."
But, in all fairness, SMP and RAID are probably pretty weak points for Linux right now, so besides the untweak they did with Samba, they did pick the right part of linux to tear apart.
Of course, what really matters, is when NT 2000 ships, will it be able to compete with Linux then?
It should be interesting to see who can actually tweak their systems more in the year (or two, or three...) to come.
OK, this is a little late in the game here (as far as I am posting a day late) but the guy did get some good feedback if you look at the thread of comp.os.linux.networking. There was a request for more info which he did not honor.
I've always noticed that help using Linux is very iterative...seldom is anything fixed on the first post.
Also notice that there was no mention of Samba, which it seems to me is where they could have used the most help.
At least his posts were semi-informative of the problem though.
Third troll in a row! Not a bad job eh, Abigail? ;-)?
Would you feel as unbiased and cool if they "benchmarked" and, say, Perl
Information wants to be beer, or something like that.
Where I work, our servers consist of:
1. Dual 300 PII, file and print server
2. dual 350 PII, Exchange server
3. Pent 133, Proxy and web server.
4. It only goes down from here.
My point? Not every company buys $60,000 quad xeon servers. In the ZD Lab tests showing linux being faster on normal hardware, attainable by normal companies. That's more important to me right now than benchmarks of huge machines I won't get for another couple years.
Ya Know?
-=nft=-
"We must be the change we wish to see in the world." -Gandhi
nt
As of this morning at hot100.com, slashdot is ranking at #28. Pretty impressive for a single machine web site.
--
While we do need to fight back, please do not slur any company names. Slurring the competitions trade names just make our position look weak and immature.
I *really* need to get moving and write the Linux advocacy feature I meant to write over a month ago....
--
Moral: Multiple network card "support" depends on the cards and the drivers. Some are easy, some are not.
--
[...]
We have Redhat 5.2 installed and compiled an SMP version of the 2.2.2 Linux kernel.
[...]
The problem: the server performs well, delivering in excess of 1300 HTTP GET requests per second. But then performance drops WAAAY off, like down to 70 connections per second. We're not swapping, the network isn't saturated (4 x 100Mbit nets), disks are hardly used, but the system is just crawling. If it were saturated then performance should level off, not drop like this. Neither vmstat nor top show anything unusual. No error messages in the web server. Its puzzling.
As pointed out by an astute poster elsewhere under this article, Linux 2.2.2 (which Mindcraft used) had a kernel bug in the buffer cache that would cause a significant loss of performance just as described above. Using a 2.2.x kernel was a little premature in this case, it seems. (Why didn't they also test 2.0.36?)
No doubt a repeat test with 2.2.5 and a proper Samba configuration would give more accurate results...
Deven
"Simple things should be simple, and complex things should be possible." - Alan Kay
Sure, 2.2.2 was listed as "stable", but common sense dictates that a shake-down period is likely to take place before it is really stable. Some bugs are subtle enough that it takes a lot of installations to bring them to light. However, many sites won't try to seriously use the new kernel versions until they are listed as "stable".
It's a chicken-and-egg problem; the kernel has to be listed as "stable" before many sites will risk it for heavy production use, and the more subtle bugs will evade detection until placed in heavy production use. Therefore, at some point, they just have to decide it's "good enough" and declare it a "stable" release, and fix whatever comes up.
Given that we don't even have a 2.3.x series yet, it should be fairly obvious that the early 2.2.x "stable" versions should still be taken with a grain of salt, for now. Yes, they're pretty stable and improving all the time. However, to expect them to be just as stable as 2.0.36, which has been in maintenance mode for years, is a bit unrealistic.
After 2.3.x kernels are rolling, and 2.2.x updates have stopped coming out for a few months, then the truly cautious sites will start to rely on 2.2.x kernels for mission-critical applications. The paranoid will stick with stable 2.0.x kernels and let the majority volunteer themselves as guinea pigs to debug the early 2.2.x kernels. (The less paranoid who can afford a small amount of instability are crucial to ironing out the little wrinkles to make a very stable kernel, as people have come to expect of Linux.)
That said, I think some mistakes were made in the 2.1.x -> 2.2.x transition. For example, major changes were introduced in the 2.2.0 "pre" series, which ought to have been strictly limited to fixing bugs. The changes to the VM system were a particular case in point here; a major overhaul of the VM system was applied in the 2.2.0 "pre" series, when it really belonged in an experimental kernel series.
The new VM code is significantly improved in general, but the bforget() bug slipped into the final 2.2.0 release and remained for several more releases until it was finally isolated and fixed. From the description of the bug and the comments in the Usenet News article attributed to Mindcraft, it seems quite clear that the Mindcraft people unknowingly crippled their Linux machine by using the 2.2.2 kernel with a serious performance bug that was yet undiscovered. No wonder the NT machine managed to win for once!
Deven
"Simple things should be simple, and complex things should be possible." - Alan Kay
Pretty much all high-traffic sites have dynamic content. They are not limited by the kind of web server performance these systems measure, but by the technology you use for generating the dynamic content (Perl, CGI, Servlets, databases, etc.).
Throughput of more than a few megabits per second is also pretty academic, at least for Internet sites and most intranet sites, simply because the network can't handle more than that.
Furthermore, Microsoft has been foremost in doing funny things with their TCP/IP implementation, both on their servers and on their clients, to look better on these kinds of benchmarks. If you look at the TCP/IP specs, it's actually impossible to achieve the kinds of hit rates they claim with a compliant implementation. Microsoft also seems to have done other things with timing and sequence in the past that made their systems look good and other systems trying to interoperate with them look bad (accident? you tell me...). So, even if NT performs better with 95/98 clients, that doesn't necessarily imply that NT is a more efficient system.
Another problem with their study is that it makes little sense to buy a four processor Xeon machine to run web sites with Linux. Four separate Linux machines are going to be more robust, easier to install, easier to maintain, perform better, and cost less. Of course, with Windows NT, because of the hassles of administering machines and because of the cost of the various software licenses involved, people may end up having to buy expensive, high-end SMP machines. I view that as a strike against NT.
They also don't seem to have tested systems where multiple, different server processes need to run on the same machine (web server, database, etc.). NT seems to perform poorly in those situations.
I can't comment as much about the Samba results. What I do know is that the Microsoft SMB servers we use seem to perform very poorly compared to the Samba servers on Solaris in practice. These are both professionally installed and maintained systems on high end hardware with hundreds of clients.
Altogether, their study strikes me as biased and meaningless. To me, NT isn't even in the running for building large, high-performance web services. For the performance characteristics and functionality that matter on real web servers, a Linux or BSD server farm is a cost effective way to go.
I have yet to read the paper, but if it comes from M$, I wouldn't be surprised to see that the computers used for the 'comparison' were different.
Does anyone have specs comparing FS performance between the filesystems supported by Linux and by NT? On the same system?
-- What you do today will cost you a day of your life.
Well, mostly to ask the /. bunch if anyone here had personal experience of comparing Linux filesystems and NT. I thought my post said as much.
When I typed it, there were no other comments posted, so it seemed like a fair question. Now, after reading the article in question, I come back to such charming company. Nice.
-- What you do today will cost you a day of your life.
There are not many things to shoot at.
:)
Linux does not appear to have done well. How does this test translate into a real world situation? Isn't Slashdot running on a lesser machine than the test server? And cranking nicely with perl and Apache doing the dirtywork?
Someone has already mentioned the 960 MB self imposed Linux RAM use limit... Looks like a typo more than anything else.
Pretty graphs that an MBA would appreciate looking at.
The testbed was purely Win95 and Win98 machines running Microsoft TCP/IP - how this translates into 'extend and embrace' is interesting.
The one major anti-Linux thing said was that documentation and support were not forthcoming for the kernel and Apache, but the Samba docs were decent. Is this because Samba is a 'clone' of a Microsoft product?
Just how intimidating is the lack of formal documentation, for an enterprise level web server? After all, the people responsible for handling such an animal would surely have readily available access to the 'routine' expertise, and quirks and oddities are not something even Microsoft documents eagerly.
Ah well.. Back to time off.
-- What you do today will cost you a day of your life.
Which is why when anything happens the site eats it. Several months back when we dropped a few bombs on Iraq (I think it was Iraq) I went to both cnn and msnbc. Sure cnn was really slow to load, but msnbc would just spit out error messages (one of which told me to change a setting on the server, giving me the recommended new value... sure give me access and I'll do it.. :). ).
BTW, cnn is running solaris.. according to at least one oneline OS sniffer site.
- My favorite error message: xscreensaver, running on an old Sparc 5 w/ 8bit color: bsod: Couldn't allocate color Blue
redhat 5.2 took longer than 98 on my machine, but that's because I was installing 400 megs of programs, while I only got a web browser and some stupid accesories with windows.
Well, it says that the NT servers peaked at *112* clients during the SMB test. Looking at the street value of the system, a 20 user pack is about $2000 CDN. 112 users would cost a company *over $10000 CDN* for just the software alone!
:)
This is just plain stupid.
æeee!
This, is a marketing move. Microsoft wants to try and nail Linux before it really begins to pose a serious threat to thier server markets.
Educated people, people who know enough about computers, are not the target of this. It's the more casual person who is going to choose the NT box when hearing it runs "twice as fast" as a linux box.
This is bad for linux users. It's really important we try and debunk this. Because if we don't, linux will stay where it is right now, a quiet underdog contented who can't get enough mainstream acceptance to really take off, like it could.
Just my take. Tell your non-geek friends, this test was a setup!
Slashdot. It's Not For Common Sense
I'm not saying it has to be totally accessable to everyone. That would be stupid and degrade the quality of the linux user community.
What microsoft fears from linux is losing the small server opportunites. Using NT to set up a web server really isn't THAT much easier thatn Linux (i had to do it, what a pain) but N$ would like everyone to think it is. For a long time that was enough, but now linux is being booked as faster and much cleaner. So out pops this test to show NT is faster.
The argument of games is totally irrelavent in this thread. We're talking about server performance of a high-end NT box vs. a Linux Box.
As I said before, tell your non-geek friends this was a strategicly intelligent marketing move. I don't need to debunk it, because this whole thread has done a pretty good job of doing that already.
Slashdot. It's Not For Common Sense
Well atleast we know win2000 will still be unstable. A friend of mine threw it on a box at school. I managed to lock it hard while trying to download netscape. Conspiricy theories aside, it had something like 3 windows of ie open and that was it. Hard drived just started chunking when I clicked on the link to download netscape and it kept chunking, after about 5 minutes of chunking w/ no screen redraws I turned the power off (since the dumb compaq boxes my school buys don't have reset buttons). The really funny thing was that while booting it did that same old chkdsk thing, then rebooted after chkdsk completed. I thought win2000 was supposed to eliminate excess reboots. Seems it introduced a new one.
-matt
I don't think this says anythign about the quality of Linux support. They claimed that they asked news groups and sought support. No one has been able to find any proof that they actually made any attempt to get help. Also, the fact that they needed help compiling the kernel, and thought that if you mess the kernel up your system needs to be reinstalled, shows that they obviously know nothing about adminstering a linux server. I think it's more a case of microsoft paying mindcraft to produce a survey that fits their needs. Microsoft has done this in the past and I'm sure they will do it again. I think what we really need is for a nuetral company to rerun the benchmark, this time allowing someone who knows how to configure a linux server to setup the linux box. Let's see how linux does then. Based on the fact that every other survey i've seen shows that linux beats nt hands down, i think that linux would prevale.
-matt
linux's SMP support at over two processors is immature in the least, so there is a considerable performance hit. I think that the mindcraft guys put the wrong line in for the amount of memory on the linux machine (append mem = 1000M or something similar instead of 1024M), some of the background programs running under linux were performing tasks that NT doesn't really ever bother with, and the configuration files were set wrong for samba.
On the other hand, the NT machine was very well-tweaked by a team of technicians who are used to dealing with NT-based (biased?) studies. The tcp/ip stack, as well as some features on the network card were also adjusted differently on the two setups. The NT raid support is also considerably more mature. All in all, it wasn't really too fair.
Linux would kill NT on a 16mb p100 or so, because all of the hardware would be well-supported, the SMP-issues wouldn't exist, the raid issue would probably not be there on a p100, and NT does not run very well at all on an older machine. You really need at LEAST 32M to make nt tolerable. More is better. Linux is fine for a number of things on 16.
Even -IF- the NT box was faster the amount data it serves will drop sharply when you factor in the amount of time it was down.
They should have done the test on Alphas. On my alpha, default 2.2.5 kernel, /proc/sys/fs/file-max: 4096. That should have solved the problem nicely. Btw, does 128 servers max really give you a benefit over 32 or 64?
They laughed at Einstein. They laughed at the Wright Brothers. But they also laughed at Bozo the Clown. -- C. Sagan
As the previous poster mentioned, /proc/sys/fs/file-max will work.
echo 8196 >
However, you may run into the per-user file handle max, so you have to
edit linux/include/linux/limits.h and linux/include/linux/fs.h and change the value of NR_OPEN and recompile. (this info from linux/Documentation/proc.txt).
Please note, linux/ indicates the root of the untarred linux source tree.
They laughed at Einstein. They laughed at the Wright Brothers. But they also laughed at Bozo the Clown. -- C. Sagan
Once Microsoft gives up "embrace and extend", one can talk about giving them credit for what they do. Until then, discussions should be about how to get them to stop "embracing and extending".
Then we can have a group positive thinking session where we talk about how mussolini made the trains run on time and microsoft made software that worked well enough for most people.
They laughed at Einstein. They laughed at the Wright Brothers. But they also laughed at Bozo the Clown. -- C. Sagan
I should also add that with $50,000 worth of hardware, no one in their right mind would be buying a Dell for a non-NT machine. Much better go with a real computer like an AS/400. I should also be noted that with that kind of hardware, you would run GNU/Linux on a Sun machine, and a $50,000 Sun machine has a much better I/O interface that would speed up intensive file serving (which is all this is--glorified file serving).
It would have been far more informative to see how well it could handle 100 simultaneous clients using a perl script of the complexity of (for example) the Slashdot engine. *That* would be an interesting comparison of Apache/Linux/GNU and Windows/IIS.
--jon. Postel is dead. May we all mourn his, and our, loss.
I've never seen the /. effect happen from a page with just a casual mention as a tagline before. Interesting.
--jon. Postel is dead. May we all mourn his, and our, loss.
X-Http-User-Agent: Mozilla/4.0 (compatible; MSIE 5.0b2; Windows NT 5.0; MS ITG IE5 internal)
That's IE5 running on Windows 2000. Interesting. I wonder what OS was *really* on the PowerEdge?
--jon. Postel is dead. May we all mourn his, and our, loss.
One of the users who answered Mindcraft's questions to ``various Linux and Apache newsgroups'' told 'em to use FreeBSD if all they wanted was to see how many files they could download from Apache per second. Of course, they didn't do that. They're missing a basic tenet of GNU/Linux: you use what works best; a webserver for serving static images would obviously be best with FreeBSD, not Linux.
--jon. Postel is dead. May we all mourn his, and our, loss.
The net posts asking for help that are mentioned in the white paper appear to have been most likely made under the pseudonym:
will@whistlingfish.net
Use DejaNews.
No-one seems to have done that and talked about it. I did; h ere's the relevant link that lists all the messages from this guy on Usenet. Take a look at them and post what you think about them. It seems to me he hit a strange, obscure bug in GNU, Linux, or Apache, and it might have something to do with network adapter or SCSI adapter problems.
--jon. Postel is dead. May we all mourn his, and our, loss.
actually it was Benjamin Disraely (sp)
one of the uk's old prime ministers
who i am for some reason I am named after...
:)
"sometimes I wish I was blind I thought I saw a whole lot more than this"
>I notice it takes 10000 requests per child.. In
>the past, a lower figure was recommended in case
>of any leaks. I wonder if that could be a
>factor.
I was left with the same impression. Not being a real Linux/Apache wiz, I can't make any definite determinations by my lonesome, but don't those apache conf settings spell death to the server?
I mean isn't 10,000 waaaaaaayyy to high for MaxReq and 1 too low for MinSpare?
Ahh well...it's propaganda, definitely not worthy of expending a lot of energy on...
"Although we may build the technology that we define as tools, we must be vigilant that those tools do not define us."
Better yet, view that same thread and see what Bill Clark's response was - perfectly helpful in relation to the questions posed...
agh...
"Although we may build the technology that we define as tools, we must be vigilant that those tools do not define us."
"Isn't Slashdot running on a lesser machine than the test server? And cranking nicely with perl and Apache doing the dirtywork? "
/dev/null I only get 1.5 MiB/s. When I did a test, for comparision only, with smbfs, I got 3MiB/s. This is with the user-level nfsd since knfsd won't compile on my 2.2.x/glibc system.
Well, Slashdot has no load compared to the benchmark they did. I mean, 1000 requests per second is 82M accesser per day. Not many sites even come close to that.
Even if their test results ARE correct why the heck cares? I mean, if someone has a site that gets that many requests, would they really just use one computer, no matter the OS? I really doubt it...
One thing though - Linux + NFS sucks. Badly. I just upgraded to a 100 Mbps network and when copying a large file from nfs (server is PII-450, raided disk) to
Well, you're wrong. The default _per_process_ max is 1024. The default system-max is 4096. So I really doubt it's an FD-problem.
Well, from a small test I just did, it doesn't seem to be the case. I made a program that forked 10 times, and then opened files in each child until it failed. I easily opened more than 1024 files totally.
I have seen a 486/Dx 33 with 16 megs of ram and 2 gig disk act as a DNS / mail server. There are aprox 1200 users at the site. The box is running linux and the 1.2.13 kernel. It has had uptimes of 40 - 50 days (until someone knocked a scsi cable off.) Lets see NT do all of that on a 486... There have been tests in the past by other companies that say the opposite. Linux is faster then NT. I think IBM or SUN should conduct the study. Not an NT shop.
Microsoft aggravates my tourettes syndrome.
Saw this on Linuxtoday.com. These guys were paid by MS to do this :) They suck!
FUD FUD FUD!
BUY my cd at http://www.mp3.com/pedophagia
Natas of
-=Pedophagia=-
http://www.mp3.com/pedophagia
Also Admin of
http://loki.linuxgames.com
These tricks probably need to be documented somewhere.
As Linux becomes used for bigger jobs in buisness a high quality Kernal Tuning HOWTO would be good. Even if it were a published book.
or better yet, skip it a bit and just get to the "DIE" bit.
"Computers will never truly be free until the last windows user is strangled with the entrails of the last mac user."
Probably, i suspect my old 486 DX2 66 is faster then that quad processor NT server =P
Hell, if you disabled SMP support the damn linux box would still probably beat the NT box.
"Computers will never truly be free until the last windows user is strangled with the entrails of the last mac user."
God, are you serious!?!? I didn't read all the details of the config. This is NUTS!
I had to go back and re-read this. I don't know how these test clients behave, but if they all pile on real quick-like, it could be a problem with only 1 minspare. You nail it with #2,3,4,5,6,... 100 all at once, and it has to first get them all loaded. Meanwhile it starts to backlog and you have to dig yourself out from under. I don't know why they would have done THAT - except to make it look bad for linux.
Also, MaxClients of 290 is probably shooting them in the foot. It looks like alot, but with 4G of ram it starts to look small. You gotta just keep jacking it up until the machine starts to *swap just a little* under maximum load. I've got all kinds of crazy modules, and my server weighs in at only 3.4M. Take out Perl and you probably loose alot of weight, not to mention all those other modules they took out. I think they should have had maxclients running somewhere around 3-4K, not 300.
Sigh, I don't even know why I bother.. *we* all know it's BS.
-=Julian=-
And finally, YES, f--ing 10000 is way too high for maxreq. I haven't checked recently, but I know I used to see leaks in apache (gasp, yes I know it's true - leaks). Cycling them is an evil solution to the problem, but a necessary one.
slashdot the hell out of mindcraft
see how long their server stands
There were a number of small things in the test that disturbed me: //No comment needed
1. They compiled their own kernel. Whether through inclusion of unnecessary components, exclusion of required features, or simple misconfiguration, it's unlikely that a bunch of testers got kernel compilation right the first time. Wouldn't you choose a system that didn't have a precompiled source system if you were going to hide some fishy results?
2. The NT server had more RAM in its configuration. It was only 44MB more, but on peak-load, it can make a diff.
3. MinSpareServers=1
4. ALL the boxes were running Windows
5. It was not a LIVE network. Only Benchmarking was used. If they'd set it as a web server, or used STMP, maybe we'd be laughing from the other side of the fence
6. Normally, I'd say M$ sponsoring is not that big of a deal, except for the antitrust reports that they have told companies what they want to see in the tests.
I would say that the Network Card is an issue, and certainly it would affect performance, but let's face one thing: The 3Com cards are bought more often than any other, despite crappy, buggy performance even on windows boxes (in my experience). They're out there and people WILL try to use them. If this is really a bottleneck to performance, then someone should really do something about the drivers for it.
That's it for my current rant.
"What's evil? What's good? There's no concrete standard for business ethics"
True science means that when you re-evaluate the evidence, you re-evaluate your faith.
They told NT things like "set maxmem=1024" in boot.ini, but did NOT tell Linux "RAM=1024" in /etc/lilo.conf. So Linux only detected 960MB.
And there were others! Which tells me these guys were NOT very familiar with Linux (like compiling the kernel). It does not look like they compiled it for the PII either. And, BY THEIR OWN ADMISSION, the RAID driver was beta. Is it binary only? I hate to be a skeptic, but we could have OEMs purposely botching binary-only Linux drivers (it would be stupid, but NOT beyond what M$ is capable of).
Also, try going past 144 users and NT will start to choke. Why? Because it needs 1GB of RAM to support 144 users. Slim the memory down to 256MB of RAM and see what happens.
-- Bryan "TheBS" Smith
Independent Author, Consultant and Trainer
The CPUs used were Slot2 Pentium II Xeons. Memory limitation is the same as caching limitation.
Early Pentium II's only cached 384MB, later ones cached only 512MB. The limitation has been removed in most newer PIIs.
-- Bryan "TheBS" Smith
Independent Author, Consultant and Trainer
FreeBSD can sign a NDA with a vendor to get specs. They did this back when Adaptec did NOT support Linux. Hence, why I used FreeBSD for a small stint.
-- Bryan "TheBS" Smith
Independent Author, Consultant and Trainer
This is no better than MS FUD, if NT could beat Linux (or for that matter any unix) on a quad processor system, then NT deserves the credit. It of course can't but that's not the point. It's about fairness.
Yeah right. My "previous" win98 (must have been
some old build or something) did not support
my 3c509b, which was supported by linux, solaris,
freebsd and (after a little patching) BeOS.
These results are exactly what any person who
understands both NT and Linux would expect.
The problem is the test was not set up to test Linux/Apache vs. NT/IIS, it was setup to test Linux/Apache vs. NT/IIS on the Dell PowerEdge 6300. They were not thorough in configuring optimal systems for both setups. An optimal Linux
solution would include more, less expensive machines in a cooperative network, not one hulking monster.
-Rich
Some people (who haven't been brainwashed by MS) actually like the ability to tune these settings the way they want to.. One of the reasons that a lot of people like Linux is because they can actually control things such as this, instead of having some "smart" code that thinks it knows better than a real live human being...
---- Dave
Very simple.
The Dell Poweredge uses a 3C905(b) ethernet chip.
The Linux driver sucks qua performance.
Hmm, very simple.
First, they didn't compile SMP support in.
Second, the Dell server they've used has a buildin ethernetchip from 3COM.
This chip (3C905b) isn't very good, and the Linuxdriver can't get more than 3,5 Mbyte/sec out of this chip.
If they had used a PCI-card with an Digital 2114* the results for the Linuxserver would been 1,5 times better than those fot the NT server.
Hmm, very simple.
First, they didn't compile SMP support in.
Second, the Dell server they've used has a buildin ethernetchip from 3COM.
This chip (3C905b) isn't very good, and the Linuxdriver can't get more than 3,5 Mbyte/sec out of this chip.
If they had used a PCI-card with an Digital 2114* the results for the Linuxserver would been 1,5 times better than those for the NT server.
Note that the Linux box was configured to only
use 960MB of the 4GBs of memory. (also note
that this was a SMP test also some specific
RAID device). Also they were testing SAMBA.
So in otherwords, they incorrectly configured
the machine to only use less than 1/3rd of
the machines memory.
I am a little concerned of specific RAID devices
also. And I personally turn off SAMBA, since
I don't support MS products (which I would consider potential security issues). (No offense
to the SAMBA folks).
Those were just the main obvious concerns and/or
misconfigurations.
Mark
Your suggestion dosen't make much sense from RedHat's standpoint. Optimizing someting means you have a target in mind. You can't optimize a box so it will be ideally configured for everyone. You optimize it to suit it's particular function. My experience tells me that Linux is mostly installed on old boxes and as far as servers go, it's small departmental servers, running a few services (web, samba, ftp, mail, dns), not usually Quad Xeon 1GB RAM boxes. Usually when you spend over 20K on a box, you'd ask your vendors how to best configure the box to fit you particular needs. Not just send an e-mail to tech support.
;)
As far as 2.2 goes, RedHat has released RPMS and mirrors source for 2.2 series kernels. They include detailed instructions on how to upgrade, and what packages should be upgraded beforehand. The point out that they do not support the 2.2 kernel because the feel it's not ready for production boxes yet. When they feel it reaches that point they'll just rev from 5.2 to 6.0 or whatever. No biggie. As a corporation putting your reputation behind their products, it makes sense to only release products that you believe are reliable, unless of course your CEO is the richest man in the world. Then you can ship dogshit
-earl
Hey slashdoters, How 'bout we bet Microsoft 1 Million Dollars they can't beat our best benchmarks on the same box? We'll build a box just like the one in the test, an we'll get the Samba team, and the kernel hackers, and the apache team, and the rest to fine tune the hell out of it. If MS can build an identical NT box that can come within 90% of the speed of the Linux Box, We'll write em a check for US $1 million.
Everybody puts in $40 (25,000 slashdotters? x 40 = 1,000,000) or so, and Rob coordinates the whole thing. In the end the box becomes the new slashdot. The Press will eat it up, and MS will lose for sure.
Or, even better, we bet em a million, and we each have to put 500,000 in the pot. Loser takes home the cash. If we end up being too poor, we can call Larry Ellison, He'll gladly put up the dough to drag MS's name through the dirt.
-earl
If you had read the first post, you would have noticed that your just repeating what he said.
Isn't this the sendfile() call? I remember the idea being taken from NT and gleefully added to 2.1.x some time ago (yearish). But though its in 2.2.5 kernel headers, it seems we have to wait for glibc 2.1 before programs like Apache can exploit it easily.
Reading this reminded me to go to the Samba website and get 2.0.3. While there, I trawled through their press-releases. They cite this ZDNet/Sm@rt Reseller article that shows quite the opposite result. They provide very little set up info for their test procedure, but they show Linux/Samba outperforming NT consistently, in a test bed with 32 clients, measured by the same benchmarking software - NetBench 5. I don't know about the Sm@rt Reseller test, but I do think that Microsoft told these guys what to find - they have done it before - and these guys then looked for configurations to support those 'findings'.
I'm wracking my brain trying to remember, Mindcraft sounds familiar for some reason, like they ran other studies and benchmarks for Microsoft in the past, and were poo-pooed even back then for their biased tests... does anyone recall this?
I'd also be very interested to see benchmarks comparing operating systems straight out of the box, untuned. Then a race between the Linux gurus and the NT gurus to tune their machines to the point they believe the performance will be optimum. *Then* compare the tweaked systems.
The network cards were Intel EtherPro 100Mbps, 4 of them. It is listed at the end of the report.
Looking at the Linux kernel source, this particular card should have good support. I'm sure that the driver is fine and performance was great... I'm also bet they only used one of the four NICs in Linux...
---- Gee, I wish I could read Kanji...
So they have 4 network cards in that nice little Dell machine, and they specifically mention "Used the affinity tool to bind one NIC to each CPU", so my question is whether they even bothered to use the other 3 NICs under Linux.
It seems to me that Linux with one network card doing only 2.5 times under NT with 4 network cards sounds about right. Give Linux 4 network cards and you get performace that easily blows NT away.
To use multiply NICs you have to use the network driver as a module I believe, did they bother to do that? I can't imagine Red Hat's installer "How many network cards do you have, but then again maybe it does, I'm a Slackware kinda person...
---- Gee, I wish I could read Kanji...
I'm sorry, but I've installed a lot of systems, and in general, with 95/98/NT/00 I end up having to go to the manufacturer's site (or use the included floppy if, on the rare case, it's not currupted). With Linux, since 2.0.x, I haven't come across a card that hasn't been supported out of the box. (Okay, there was the GB card, but 95/NT didn't even know what it was). I'm even including notebooks (PCMCIA/PC Card) and Compaq 's (with those built-in TLAN cards, or more recently tweaked Intel cards). I'm sorry, but my experience is not the same as yours. (Trust me, I get REAL annoyed when I get NT installed and realize it doesn't like the Network card, and the network drivers from the site are about 2k bigger than a floppy).
-- Keith Moore
This sig is the express property of someone.
The beowulf newsgroup had a couple short threads a couple months ago about consistently abyssmal performance on redhat 5.2 SMP machines running 2.x with > 512 MB of RAM. The two threads [ one, two] deal with users who had horrendous performance problems with their new machines (both running 2.2.2, the same kernel as in the report) when they used more than 512 MB of ram, but the performance jumped right back up when they used 512 or less. Check out the articles to see how bad the performance was; it's pretty surprising, and presents an interesting opportunity for detractors of linux:
Linux definitely has some hardware/kernel combinations that would seem OK by design on paper, but exhibit peculiar behavior in practice, especially with SMP. I wouldn't rule out the possibility of the testers (or financial backers) hand-picking kernels/hardware configurations that could affect results while seeming perfectly viable to the layman.
It seems very likely to me that if Microsoft did not outwardly donate the hardware to the testing company, they at least made suggestions on its configuration. The open nature of linux development and bug disclosure could easily be used by companies wishing to stage biased demonstrations; Microsoft almost certainly does a thorough job tracking linux kernel development and bug reports.
-- Scott
Try google:
www.google.com
I like it, it's got a cool name.
I've seen enough charts and statistics to know that one can almost get anything to look good if the right/wrong questions are asked.
I am ignorant of Linux, running WinNT. Having admitted that, I wonder, barring outright lying, what are some possible explanations for these results?
They use a Quad Xeon 400MHz system. Is there any problems with this? Does Linux support 4way SMP well? Does it support the Xeon well?
Would it, in otherwords, be more fair to test this on a single Xeon 400MHz system? Unless I get responses to the fact that Linux does indeed have robust and reliable 4way SMP... I was under the impression that it was a new thing..
It just seems, unless Linux is 4way SMP capable, that they just ran a test of a 4way system(NT) vs a 1way system(Linux)... Hopefully someone will give me a correct answer.
Another thing that occurs to me is the use of 8HD raid for data storage...
What is Linux's benchmark there?
What if mindcraft had run the tests on a P100 with only 16mb of memory? Who would have succeeded then? My point being(unless Linux really does suck), the choice of the hardware is as important in the high end as the low end.
They could have compared the two OSes across a range of hardware, rather than this specific setup, to see which one is the best solution for a setup, excepting that they were paid by M$ to run this test in the first place.
AS
-AS
*Pikachu*
Use THOSE statistics. Until then, would somebody tell those idiots that there aren't that man all-Microsoft IT shops in the world. Convince them that there is no such thing as one tool for all jobs. Each OS does a specific function best. (Or would you rather work on your car with nothing buy a Leatherman?)
"If Microsoft were a religion, he'd be the Spanish Inquisition." - Me, about a don't-do-it-with-anything-other-than-MS fanatic at work.
The article says MS "reach a peak of 3,771 requests/second and (Mbps) of throughput."
In the real world, bandwidth costs money!
Hmmm let's see... how many web sites you know have pipes to the Internet that can handle that?? How many T3s is that?
Another way to put 22.4 Megabits/second in perspective is that this load figure is in excess of just about every major web site except the big portals.
Yahoo.com (167 million pages/day)
ie (((167,000,000 / 24)/60)/60) = 1932 (approx)
so they are serving 2000pages/sec, NT can handle 3771 req/sec (pages I assume)
Damn even Yahoo dosen't need the system they are running, sure like to meet the people that do!
If the "wide links = no" was set correctly as others have said, maybe Linux might have reached that 2000pages/sec
If I were to do a benchmark, I would publish the specifications of the system.. i.e. Quad Xeon, 1GB RAM, RAID 5 etc, specify this to the hardware adapters to make sure there is no hardware based tuning that was done. Then I would approach the hardware manufacturers who would LOVE to jump at a case like this.
Microsoft could designate one, and RedHat could designate another. After all, when one wants to go to a large server based installation, a corporate MIS person, would spend some time researching the hardware vendors, and select one that had a good understanding of the Operating System selected and could support them.
I would then publish the benchmark to the two hardware vendors and suggest to them that the pre-installed operating system on their hardware should meet certain security and auditing requirements. This would eliminate any perverse software fine tuning and would simulate what a corporate requirement would be. If the goal is to simply judge an operating system by its benchmarked performance and not by its true value, so be it, let the respective vendors who know the hardware and software crank away at the settings, as long as they do not violate the base requirements set out by the benchmarking company.
This is the type of benchmark I would trust.
The benches are not worth of slashdotters eyes. First, they describe PEAK PERFORMANCE in EACH chart. What about sustained/average rates??? Second, why don't we all get together and sue (woohoo, another Class Action Suit) M$/Mindcraft for false advertising/slander against Linux (AFTER we find somebody willing to buy a copy of NT and Linux and test it openly and independently). Third, who the hell is that Wonko42 guy? his site, http://wonko.com, shows just how current he is. An attemp at /.ing that fails miserably (and turns my stomach, too).
This is my sig. There are many like it, but this one is mine.
Can I have some of your crack? Because there is NO retail or OEM version of Windows 98 without support for 3C509 series of cards. (I used to earn a living configuring Windows 98 for networks - sad but true). The only possible explanation is that you were working with a beta copy of '98. I find it interesting that you fail to mention this fact. You are either intellectually dishonest or ignorant of the situation. I doubt that it is the latter. I hereby dub thee a FUDster.
So, can I borrow $905 from you? I will pay back the $509 I owe you within an hour. :)
.INF file doesn't contain accurate information on file locations. You almost have to install the driver files by hand.
I was addressing the 3C509 (as was the post I replied to), not the 3C905. The 3C905 situation is just insane. There is no possible way that an inexperienced non-geek could install that card - the card misdetects or misinstalls and the
Linux is just getting into SMP and probably needs more tuning. Though I'm pretty sure they didn't do their own tuning other than replacing the kernel.
Doesn't somebody make a special Server Edition for this sort of thing though? I though it was RedHat but I can't find it... If there is, try running the test with that and see what happens.
Glückwünsche, haben Sie Slashdot ermordet, indem Sie zum korporativen Druck beugten und Subskriptionen einlei
I would be careful as to what this returns. Our web server is behind the firewall and it told us we were running IIS 4.0 on BSD/OS. heheheheh. I didnt know there was a port ;)
"Fighting the underpants gnomes since 1998!" "Bruce Schneier knows the state of schroedinger's cat"
I wonder if this was done by the same people who did the infamous video in the antitrust case. There are four things to keep in mind with this test:
1) It was produced and funded by Microsoft.
2) Benchmarks only measure a few factors and don't neccessarily translate into real world results.
3) Optimization is an art as much as it is a science. Microsoft people are probably better at optimizing NT than Linux.
4) Can the results be duplicated by independent parties.
Since the results go contrary to most of the other throughput test, the results probably are not valid for one of the reasons listed above.
In a world that is Free and Open, who needs Windows and Gates?
they were using a ZD program to test SMB performance?
So? ZDBOp's own tests on Linux (using lowly 2.0.36) showed that it was almost twice fast as NT. In fact, the headline of the article was Linux Up Close: Time To Switch. Unfortunately, this version has pulled the damning charts.
Mike
--
Mike
--
"Wi nøt trei a høliday in Sweden this yër?"
They act as though frequent updates are a Bad Thing. As if hoping and praying that the next NT SP will fix your problem, be released within the next year, and be affordable were better...
grrr...
Mike
--
Mike
--
"Wi nøt trei a høliday in Sweden this yër?"
I can't imagine Red Hat's installer "How many network cards do you have, but then again maybe it does, I'm a Slackware kinda person...
I can assure you that it does not. I have two: 3Com 3c900 (PCI, 10BaseT) and 3c503 (ISA, 10Base2 -- don't laugh; I got four of 'em for free!). It detected and properly set up the 900, but I was left to my own devices with the 503; RH found the one and assumed that it was all I had. These biased and clueless newbies may not have even been using the other NICs at all.
Mike
--
Mike
--
"Wi nøt trei a høliday in Sweden this yër?"
If Apache is the bottleneck, I might be able to comment on why:
AFAIK, Apache creates multiple heavyweight processes to handle inbound connections. I'm not sure of the ratio of processes-per-connection, but the overhead of this concurrency model is rather high.
Higher performance web servers (like Zeus) utilize a thread pool (server threads stick around rather than thread-per-request) model along with certain asychronous IO features (I think) to speed up performance. I'm not sure of Linux's asynchronous IO support, but apparently the underlying filesystem is large bottleneck of any server.
I could be talking out of my rectum, so any Apache gurus feel free to flame me, but this was my understanding of the problem.
If you want to read more written by people far smarter than I, see
http://www.cs.wustl.edu/~jxh/research/
There's more information there.
-- "I am disrespectful to dirt. Can you not see that I am serious!"
I suspect that Apache's number 1 status is due at least partly to the fact that is comes free with many unix flavors. There are better packages around. I posted another message to this end, but I thought I might give another link to a benchmarking paper put out by my research group. It may not be the most recent, but AFAIK, it's still accurate.
-- "I am disrespectful to dirt. Can you not see that I am serious!"
Perhaps that is part of the problem, but I think my original point is still at least partly vaild. The overhead of a "process-pool" is greater than that of a thread-pool. Threads can share many more resources and thus context switches between threads (although unavoidable) takes less time than context switches between processes (Silberschatz/Galvin). But yes, the pool of processes/threads should be set to a appropriate size relative to the expected load.
-- "I am disrespectful to dirt. Can you not see that I am serious!"
Even if NT is a faster file and HTTP server, you have to realize that NT is very unstable, while Linux is one of the most stable operating systems. Even if Linux + Apache is slower than NT, you don't want your file or web server going down often. A Linux based server may be slower, but it won't go down often, if at all. NT has a horrible track record of being unstable, while Linux has a great track record of being very stable.
This may explain why many sysops prefer NT over Linux or Uniz.
DES Khaddafi KGB genetic jihad Uzi Rule Psix Qaddafi cryptographic Peking Mossad Legion of Doom Albanian Serbian Saddam
That could be true, but I have an idea that they did not set up SMP properly, or have it running at all. Even if I felt that NT could beat the Linux configuration they had (set up properly)-there is no way it would beat it by such huge standings. If NT was utilizing all four processors, and Linux only 2 this would be feasible. Now all they need to do is have a benchmark with a constant load, just to see who goes down first ;-)-by the way, I didn't see any mention of how many times the NT server crashed when it reached it's peak performance all those times, did you?
http://www.netcraft.com/cgi-bin/Survey/whats?host= www.hotmail.com
IMHO, one of the things that slows NT down (and, for that matter, Solaris too), is that it uses a number of abstraction layers in its 'kernel'. Sure, it makes for a more elegant kernel structure, but at severe cost to speed.
:)
(Solaris, apparently, is a dream to write drivers for, but is around half as slow as linux. NT's still a bitch, but that's M$ for you
If it encounters errors in the NTFS (and perhaps even a DOSFS), then it needs to reboot after its corrected the error. In all fairness, even Linux does this.
Rather than prove NT is faster than Linux,
it proves MINDCRAFT's engineers have zero
skills.
This should cast severe doubts on MINDCRAFT's
ability to do any benchmarks.
Objective trade press should take note of this
for the future. How much is your credibility
worth MINDCRAFT ? Was it worth what MS paid you
to do the "study" ?
Nitin Borwankar
--------
The more idiot-proof you make it the smarter the idiots get.
Considering they claim that they configured the
OS's equally, which they didn't RedHat and the Apache group should sue MINDCRAFT and as a
part of the process demand to see all the
documents defining what Microsoft asked MINDCRAFT
to do. These should be made public.
Nitin Borwankar
------------------
The more idiot-proof you make it the smarter the idiots get.
If you check their main whitepapers page you will notice that they did two similar studies previously, one between Netware and NT and one between Solaris and NT. In both cases they compared the price/performance of the OSes involved. They didn't this time. Might I conclude that thay have a selective observation?
Hamal is an yellow star in the constallation Aries.
It is 66ly away, so it doesn't alter your personality.
But, does "set maxmem" actually do anything? It looks good and all, but how do we know the OS really pays any attention to it?
I'd be happier if they'd physically yanked the extra RAM out of the box. Why didn't they?
-- Alastair
That message stated that they were running "the latest 2.0.3 version of Apache." The report has them running 2.0.3 Samba, and 1.3.4 of Apache.
It doesn't help, if you're trying to get an answer, to confuse your audience. (Although I'm sure this was unintentional.) It's interesting that one (only?) reply told him to "stick with 1.3.4", but did ask some follow-up questions. If will@whistlingfish ever replied, it was not via the newsgroup.
As someone else said, they didn't try very hard.
-- Alastair
Now, here's the complete reply they got.
In article ,
will@whistlingfish.net wrote:
> We're considering using Linux + Apache as a web server.
Excellent choice.
> The hardware is a 4-processor 400 MHz (Xeon) server with 1GB of ram, a RAID
> controller, and six disks. We have Redhat 5.2 installed and compiled an SMP
> version of the 2.2.2 Linux kernel.
I hope it's not too late to change your hardware, because your box is a
complete waste of money. SMP gives you *nothing* with regards to web
serving, and it makes your OS flaky as all hell. The RAM is nice, but the
processor speed is overkill and having 4 of them is just plain wasteful. The
network card would saturate completely before you even came remotely close to
using up the resources of even a single P2 200Mhz.
> For the web server we used the latest 2.0.3 version of Apache.
Stick with what works. I'd use 1.3.4, as it's generally considered more
'stable'. You don't *always* want to be "bleeding edge".
> The scenario: we're bangin' on this web server with a bunch of clients
> to try and get a handle on its capacity. Simple static HTML requests,
> no heavy CGI yet.
Another suggestion: mod_php3. I guarantee that if you ever see large
amounts of traffic, CGI will rapidly become your worst nightmare. There are
a variety of _internal_ Apache modules that give you everthing CGI can do,
but faster better and more efficiently. Keep in mind that CGI requires you
to fork() another process to handle each web request, which can very quickly
run you up against the process limit on a heavily loaded machine. PHP3 is a
PERL-like, C-like programming language that's relatively lightweight. You
can download the sources from http://www.php.net/, where they also provide
instructions on how to build it into Apache.
> The problem: the server performs well, delivering in excess of 1300
> HTTP GET requests per second. But then performance just drops WAAAY
> off, like down to 70 connections per second. We're not swapping,
> the network isn't saturated (4 x 100Mbit nets), disks are hardly used,
> but the system is just crawling. Neither vmstat nor top show anything
> unusual. No error messages in the web server. Its puzzling.
Try various flags to netstat, see what they say. If you could post the
details of several different commands that would be helpful in diagnosing the
problem.
> Any ideas? Any tips, suggestions, or pointers would be appreciated.
> Thanks!
What type of network load do you expect to see on your box in the long run?
What type of applications does it need to run (other than Apache and its
modules)? I know it's blasphemy in this group, but if you're just doing "raw"
webserving (no database interaction) you'd see *much* better performance with
some variant of BSD (for example, FreeBSD from http://www.freebsd.org). If
you're more into running a K-rAd k00l website with lots of doo-dads and gizmos
(and don't care about performance under heavy load), then Linux is your best
bet.
-Bill Clark
-- Alastair
Maybe they forgot to enable TCP/IP support when they compiled the 2.2.2?
=D
--
Okay, I got Linux installed. So where's the free beer everyone keeps talking about??
They also said WoNT 4 outperformed a Sun server (forget what it was running, or even what it was), and had a 10x superior price/performance ratio.
--
Okay, I got Linux installed. So where's the free beer everyone keeps talking about??
The white paper says:
We posted notices on various Linux and Apache newsgroups and received no relevant responses. Also, we searched the various Linux and Apache knowledge bases on the Web and found nothing that we could use to improve the performance we were observing.
Anybody see these? Somebody do a dejanews search.
Btw. Is is possible to do all the NIC tweaks they describe with the Intel 100 linux drivers?
Anybody know what this does:
Used the affinity tool to bind one NIC to each CPU
(ftp://ftp.microsof t.com/bussys/winnt/winnt-public/tools/affinity/)
If this does what I THINK it does it would explain a lot.
First, I don't care whether Samba beats NT on fileserving. I'm not a big Samba user and won't claim to comment on it, but I do know something about Apache:
:}
First, I noticed that Apache falls over when the number of requests reaches exactly half of the MaxSpareServers setting, and that they had a hard maximum servers compiled in of 500. Obviously somebody knew something about Apache; you have to pay attention to set that value.
I would like to see the results of the survey with different MaxSpareServers value. Also, Apache does not spool up terribly fast, even with the new faster preforker. With a StartServers of 10, it will take it a couple minutes to reach operating temperature. The Apache team mentions in the documentation that this is part of why Apache is so much better than the benchmarks make it seem. The test could be over by the time it's getting fired up.
I don't think the switching overhead was the big hurt here. That system is MORE than fast enough to handle a few hundred simultaneous connections.
Another interesting thing would be for them to take this exact same disk drive and plug it into a much smaller box - say a dual Pentium 133, or something. I wonder if maybe it wouldn't generate nearly identical results, thereby demonstrating that configuration is in fact the issue.
And, of course, Microsoft paid for the study. It looks like the lesson of the Halloween memo may have been overlooked at MS and they are going to try to use FUD after all. That would be a colossally stupid thing for Microsoft to do and would make me very happy
Oh for crying out loud. The FreeBSD people are still quoting problems with Linux dating back to 1994. Linux' network code has been fixed for two, almost three years now.
The readlink() function tells you both whether a given file is a symlink and where the symlink goes, so testing whether something is a symlink could indeed be a possibility.
No, it doesn't recurse. After some more thought on this, I think a chdir() then a getcwd() really is the way to go about it (I'm still trying to figure out why you need three of them, though...).
Say I put /home/samba on as a public SMB share. Also say that there is a symlink called "root" in this directory that points to /. Then someone accessing this share can "cd root" to access the whole filesystem. Setting "wide links = no" prevents this by causing samba to check if symlinks are outside the share before following them. However, NT4 doesn't have symlinks at all anyway, so in comparing the two, it is acceptable to just delete all symlinks from SMB shares (or at least all those that point outside the share).
I'd prefer Linux any day of the week precisely because of it's own credit. I mean it doesn't need the FreeBSD's or any other *nix' one. Microsoft not using it's own NT proves its inferiority but it doesn't by itself prove Linux' superiority.
It's evident though whose side you aren't on.
Why is anyone surprised at this? Microsoft attacks anything that is a threat. Microsoft has always attacked a strong competitor with overwhelming market force. It will consistently produce more evidence of how poorly it's competitors' products 'truly' stands against its products.
Their corporate methodology has been clear since their beginning. They assume the competition's strengths into their products, and then they state that their product is then the superior product. Their tactics have always been obvious and simplistic, and their tactics have also been very effective, until now.
They face a quandary with Linux: how does a business compete with a product that has no specific vendor to attack? How do they compete with a product that is communistic in nature? It is more than their product's competitor; it is becoming their corporate nemesis. They cannot overwhelm something that has no boundaries, that is developed without regard for specific profit, and that has their own corporate policy as it's core design: 'Be the Borg.' -- take the best of your competition's strategies and products, and make it part of your own structure. Anyone who has followed the history of the computer's evolution will remember that Microsoft started as a forced progression of business policy into non-mainframe OS software development in the late 70's. Anyone who remembers the early days of the PC (or then known as microcomputer) will remember that the thought of 'licensed' software that was property of only the company that created it was laughed at initially as unworkable or unsustainable, but Microsoft succeeded in making that policy work. Microsoft grew rich on that one idea, and it is the reason that Microsoft has been able to achieve dominance in the OS arena.
BUT, Linux has changed something important, and Linus probably didn't realize how important what he did was at that time, or what part of it was important. Linux by itself would never have had the possibility of competing against any dominant OS. It would have been another hacked OS that would never have left the collegiate world. It IS the Open-source licensing structure that has added the needed element to the software that has made it into an upheaval in software design methodologies. It's the Open-source piece that has turned a lot of heads due to its impact on the software industry. This is due to the fact that Linux (and through Linux, the Open-source licensing structure) is an evolutionary change in software design. Linux started as a free, cooperatively evolving OS that has returned the unstructured human element to the process of business software development. Sadly, this human element has always existed in the academic community, but died in the business community with the domination of Microsoft as the dominant business model in the software industry. The corporate structure that has grownup around Linux is just a natural reaction of capitalism to anything that has the ability to produce revenue, but Linux remains a communistic product by it's licensing structure. And that's a good thing for it; it's the only way it will be able to remain a strong and vibrant competitor of Microsoft for the long term.
So, in the end, the analysis of Linux vs. Microsoft is a null argument. Microsoft cannot compete with a product that is not a product, but a movement. Linux is fundamentally restructuring corporate policy towards software development. I just hope Linux's impact will survive the greed that will try to control it's nature while the Open-source movement grows up.
And I hope the 'Borg' in Microsoft can change it's ways so that it can allow another dominant player into the game without it having to feel the need to annihilate it.
-- The violin is playing in the background for those who are listening to it too.
To dream... To fly... What I really want is to be the fly on the wall of your dreams.
uhm, how about 486's =)
that would show true NT preformance right there!
---bob
The time lost every time they had to reboot NT.
'nuff said.
The readlink() function tells you both whether a given file is a symlink and where the symlink goes, so testing whether something is a symlink could indeed be a possibility.
It would have to be called recursively, wouldn't it? A symlink might point to another symlink inside the exported area which points outside the exported area. Or does readlink() recurse?
yeah, really fun..
people at Mindcraft, please configure the NT server at www.mtnsms.com. I never get an "HTTP/1.1 Server Too Busy" from the old sms server (Netscape+Solaris) or other heavly-loaded non NT server's like yahoo, altavista or slashdot
well, my script should have hit em a couple hundred times by now. =)
GAT d- H s++:- g- p3 !au a- w++ v* C++++ UL+ P+ L++ 3 E--- N- K- W+ M-- V-- po Y+ t+ 5+++ jx R G++++ tv b+ D-- B--- e*
you are familiar with the process of lobbying are you not?
GAT d- H s++:- g- p3 !au a- w++ v* C++++ UL+ P+ L++ 3 E--- N- K- W+ M-- V-- po Y+ t+ 5+++ jx R G++++ tv b+ D-- B--- e*
i believe they're referring to the 2.0.36 kernel... as to whether or not the claim of bugs with RAID and SMP in there is legit or not, who knows... it doesn't belong in the article (the M$ funding helped it get in there though) though since they were using 2.2.2 anyway... ~shrug~
GAT d- H s++:- g- p3 !au a- w++ v* C++++ UL+ P+ L++ 3 E--- N- K- W+ M-- V-- po Y+ t+ 5+++ jx R G++++ tv b+ D-- B--- e*
Imagine their rc5 keyrate with 144 pentium 133s and a quad xeon box.
:)
By my rough estimate, that's 29,280 kkeys/s, or about 9428 blocks/day. That's some serious code pumping...
-Chris
Let's find someone at RedHat who'll be willing to actually run the test, and let's just start sending scads of hardware to him/her so the test can be duplicated, with a PROPLERLY set up (and tuned) Linux box and a properly set up NT box.
I'd pitch in my three P133 mobos. And probably cash for the server, too. This is worth it.
-Chris
Remember in the doj trial when said that billy saying that it would help him a lot if if he "had a survey of developers who wanted an internet browser integrated into the operating system" and he was able to get that.
Now we have linux making significant inroads into markets that microsoft wants NT to enter. Shall we ask bill if "he wanted a survey that suggests that linux can't handle server load on high performance machines compared to NT".
-?-
Actually, the last test I saw(performed by ZD, no less), indicated that Samba serving Win9x clients is about 3 times faster under heavy load than NT.
- Free tabletop fantasy gaming! Grey Lotus
in redhat you actually have to set up the 3 remaining network cards "by hand", the first one is the only one redhat "recognizes".
-- To bloody go where no man has gone before.
Anyone know of their hardware setup nowadays?
I recall when they used to mention it, back when they were several Pentium Linux boxes.
It would be interesting to know if they are still Linux servers, and what their load and etc is..
OK.. no one mentioned it that I've seen so far, but it kinda glared out at me: the fact that he mentions performance "in excess of 1300 req/sec"
before it falls down.. then in the report, that shrinks somehow to 1000 req/sec??
Kinda makes (me at least) ya go Hmmm...
Anyway... I'm looking forward to results on the same kind of hardware, peaked by people who know how. Hope they get that re-test project going!
Well, we have a VAR 4100d (4100c? i can never remember). Anyway... 2GB ram and 1.2TB disk space... The performance so far has been amazing, and I'm afraid to see what would happen if it had NT on it (BSOD probably). It handles all of the windows boxes in the office getting mp3's off of the samba share and the linux boxes off the an nfs export... Among it's other tasks of course. It is simply remarkable how stable and responsive the server is. But that's just my experience.
1) 4 physical gigs of ram when linux can support 2 gigs.
2) "Red Hat did not ship or support a product based on the Linux 2.2.x kernel." pardon me, but (all holy wars aside) RedHat is NOT linux per se, just a distribution. This always bothers me. The mentality of "since we couldn't buy it right off the shelf, it doesn't exist" but they are quick to use IE5 (when it was beta) and betas of office and windows 2000 when it suits them. Our kernels may be "beta" but they are actually out there.
3)"we found the following kernel update history:" Turning a strength (frequent updates) into a weakness (oh, Linux has soo many bugs they have to constantly releas) is a typical FUD tactic. It's so popular it's even in the big document defining the concept of fud. When was the last time Microsoft released a kernel?
4) "Linux performance tuning tips and tricks must be learned from documentation on the Net, newsgroups, and trial-and-error." and I wouldn't have it any other way. At least through this method they are *free*. There is a big difference between MCSE courses and common knowledge.
5) "For people not familiar with UNIX/Linux systems, it may take longer to do the installation." And the same is true for NT admins. You are hired because you know how to do something. That's the point of business. Why would you hire somebody for a Linux position if they don't know Linux?
Anyways, this entire document is sketchy at best. I submit the following configuration for test:
Quad processor Machine with 2 gb of ram (let's be honest, linux does not yet go above this afaik), Properly formatted hdds, NT4.0, Linux 2.2.5, Raid 0, HDDS formatted for NTFS and ext2fs respectively (not nfs -- maybe coda), Apache should have mod_cgi compiled in (the benchmarks had a section for this, I believe), and configure linux right this time!
btw: why is it that when I swap windows in my barely accellerated X-Windows system in rapid succession it can keep up with redraws, and with the same system in win95 it won't?
Lowmag.net
After reading all this comments I feel Linux need
a HOWTO on how to tune a system like a quad Xeon
with 1 gig of RAM and RAID to the proper
performance. This kind of box is not the tipical
hacker machine but reading the comments on this
page already showed many things that could integrate a checklist.
Maybe a new Linux distribution (or a new flavor of an existing distribution is needed - Like High End Server Debian) is needed. Despite the obvious bias in the study other thing is obvious: They are NT people, and people without experience in Linux could hardly tune a regular machine.
- So Linux does not scale well out of the box:
Solution 1: Create a packaged solution.
Solution 2: Create instructions on how to correct
the problem yourself -> the Open Source way of solving problems!
Actually, the observation is germane...Apache was ported to Linux AFTER ports to commercial server OS's (INCLUDING NT! Read the CVS trees) and by a separate sub-group. It still shows some problems with memory leakages and some consistent I/O errors under heavy load, if you implement the binaries or the standard source on your server. IMHO, the ZEUS applet web server on Ultra-SPARC machines (especially with the ability to run Java servlets, rather than Perl/other scripts environments, with their back-end CPU loading)pounds the tar out of Apache in a commercial environment BEFORE you consider SMP issues. Apache and IIS are the most used primarily because they are FREE, not for any technical reasons. It also helps to remember that all the modern servers derive most of their code from MOZILLA/Netscape, and that IIS has more than a little "cloned code" from Apachhe source. Iguess that that is one of the disadvantages of Open Source: the Enemy is free to take advantage of your work, too. Dead Mike
how about we put together a petition to mindcraft for a very biased redo of their test on their hardware. for instance, have alan cox/linus torvalds configuring the linux box versus steve ballmer/bill gates (or whoever microsoft wants) configuring the NT box. give them both exactly 3 hours (or 4, the idea being the average amount of time an administrator puts in when (s)he configures a new webserver) to fine tune everything. then re-run the benchmarks (including non-ZD ones) to determine the winner.
At a LAN party, we were serving files off a dual-boot Window-FreeBSD box. It was running slow real slow of windows, so we switched to BSD and it at least doubled in speed. Enough said.
Yeah, Mindcraft also authored a study (commissioned/sponsored by M$) that alleged to
prove that NT is faster/better than Novell NetWare. Novell claims that Mindcraft used ZD's
NetBench in ways that Ziff-Davis warned would
produce skewed results.
Apparently "independent" is a flexible term.
This looks pretty fishy. After reading everybody elses posts I think I'm gonna go w/ the opinion that it's fake.
Obviously this test was grossly biased. And if you did another one, then you'd probably get another biased, if not equally biased study. It could go either way depending on who it was doing the test, but you can't get bias free.
So why worry about bias at all? Why not just put out a challenge and find 2 very capable teams. A Linux team & an NT team. That way each will know enough to tweak their box & won't intentionally or subconsciously disable features on the OS they like least. Of course this could actually benefit NT, if some Linux people disabled 1/2 the "features" that MS packages into any of their products.
The HW could be a mutually agreed upon by both teams as could the exact tests thrown at the servers.
Certainly MS has some redeeming qualities, however their stereo types are well deserved.
After the discovery of the Bill Gates memo "requesting" results from an independent testing lab, I would think any further MS sponsered testing would be completely ignored - especially when reviewing Mindcraft's other recent "white papers" (NT faster than Netware, NT faster than Solaris 2.6... *yawn*).
Bah. Obviously these people are lamers, its almost not worth the time to comment...
I would like to propose that the linux community take the initiative to not only redo this test, but do so fairly. Take 2 identical machines, preferably the ones they used in the original test, assuming the hardware is completly linux compatible (which it probably was not), give one to microsoft, or the MindCraft. Let them configure their NT and tweak it to their hearts desire. Let a couple of linux guru's configure linux to the normal highend server standards (hey why not give them the advantage.. laugh). And run the test at a neutral location where all can watch. Hey give prizes to the winners. Let people bet on the benchmarks, whatever, would be a great event.
Their resaoning is along the lines of 'our server is fast enough on a half-decent machine to saturate a T1/T3 so there's little gain in going all out to make it go faster.'
Certainly in terms of webserver performance it isn't necessarily realistic to load a machine with four 100MBit network cards which you then hit as hard as you can, that's equivalent connectivity to 9 T3s.
But the apache attitude is 'we want our server to be able to handle the needs of all webmasters as well as we can' not 'we want our server to get good benchmarks in artificial conditions.'
There are servers out there which can outperform apache, but it's harder to compete on the sheer range of things apache can do.
-- David Mansell, Cambridge, UK.
I'd like to see this sort of test with a third, control group thrown in for good measure, such as Solaris. I personally couldn't give two hoots about NT vs anything, it's not unix which is all I want on my systems, and I'm sure there's a lot of managers out there who think the same.
NT and unix are too disparate for a balanced comparison, high-end Solaris vs Linux comparisons would offer a clearer perspective in the real world.
POKE 36879,8
It has been suggested that their problem might have been a bug in 2.2 that was fixed in 2.2.5, namely the bforget() bug.
e /reports/bforget.html
;-)
http://www.citi.umich.edu/projects/citi-netscap
I think they would have noticed the buffer cache blowing up and processes getting stuck if this were the case, though. I personally favor the theory that they left Advanced Power Management turned on in the BIOS and after 30 minutes it was putting the CPUs in powersave mode.
According to Alan Cox (and he should know) 2.2.2 has a known TCP flaw that is triggered when talking to NT servers. Apparently, this bug still exists.
:)
The Linux developers care about this issue, but not so many of them have NT running at home...
The problem comes when you start adding modules to Apache like mod_perl. This is when the forking mechanism really starts to suck. The individual processes can get large, limiting the number of concurrent connected users by memory footprint alone. An Apache fork spends most of it's time in network communication on real WANs in my experience, NOT content retrieval.
What's better is to have a smallish pool of subprocesses and then pass every client to a small, fast, single-forked buffer (an accelerator) so that clients don't sit on the big Apache processes due to slow international network bandwidth. Instead, they talk to the accelerator, which as it is very simple (a straightforward buffer) can take many clients. In effect, you're allowing many many more clients to stay connected to the server concurrently, while using a smaller number of Apache forks to actually construct and arrange the content response.
You can use something like Squid for that, but Squid can be a bit of a monster. I'd like to see a really small non-cacheing accelerator based around a single select(2) loop....?
As far as I know, the architecture for IIS is slightly better in this respect, in that the bit of the server the client listens to is effectively detached from the server thread which builds the response (the big bit).
I might be talking complete balls as I haven't had a chance to play with this structure yet, but I've heard reports that it flies. In which case, it might be a fairer test.
Very well said!
;)
I have to agree with others' posts here though - you don't need a Quad Zeon to do web serving.
Running a PII-350 here with 128Mb RAM, Linux 2.2.5, Apache 1.3.6, MySQL, PHP. Flawless performance. 'Nuff said
Don't throw your computer out the window, throw the Windows out of your computer!
I'm sorry to say this, guys, and I may be missing something, but from my experience over the last few days (I just installed NT4 because of a burning need for 3d Studio MAX :-) Apache on NT seems to actually serve up pages faster than Linux on the same box serving up the same pages. When I access the page from across campus while Linux is running, there is a somewhat noticeable delay before the page pops up (very small, though.) When NT is running, the page pops up with almost no delay. While this upsets me, there's no point in lying about it because then we'd be no better than Microsoft.
BTW both setups of Apache were un-tweaked, freshly installed copies of 1.3.6 serving up the same static pages off of their native filesystems. The server is a P2-350 with 64 mb of RAM. The NT side of things is running NT4 and the Linux side is using kernel 2.2.1.
And NO, I'm not an NT junkie. I'm a hardcore Linux fan who regrets having to run Microsoft products at all, but I just want to tell it like it is.
If anybody would like to comment on this to me, please send me an email rather than post it here, because I'm probably not gonna end up coming back to this page. (spong@isr.umd.edu)
"Software is like sex- the best is for free"
Wow.... SOMEBODY had a little too much fun with xpaint :-)
"Software is like sex- the best is for free"
In what way would it be sane to use NTFS instead of ext2 as the filesystem on the Linux box? It does seem odd that one of the system are allowed to use there native filesystem and not the other, in a I/O intensive test?
I think the 'unless they are fools' is the key here. Obviously the way they configured Apache shows their statement that "Both Linux and Windows NT Server were tuned to perform optimally under each of the two workloads." is false.
I agree with someone (sorry, I'm too lazy to check who) who stated in a previous reply that the systems should be set up by an expert provided by the vendor.
Oh, that's right, silly me... M$ funded the study, so we wouldn't want that.
There are some interesting comments regarding the
7 14f389-03702e40
test at:
http://www.segfault.org/story.phtml?mode=2&id=3
>>>>
I will have to say that I think ie4 is the best browser I have used and I have never had a problem with the MS Office products. When people bash MS it usually isn't for the entire product line, it's for the OS and their business practices.
I don't see how you can state that office2000 and win2000 and ie5 are "of exceptional quality" when they are far too new and untested by the masses. Win2000 doesn't even exist in commercial form, just beta.
They claim that they contacted RedHat for help with configuring the kernel and RedHat wouldn't help. Makes me wonder... They also said they posted in various news groups and didn't receive any help. This goes against all of my experiences with the the newsgroups. This really smells of FUD... You have to give the boys from Redmond credit, they are very good at promoting their products and FUD.
Actually, linux running SAMBA has repeatedly been tested as being a "better NT server than NT". I recall one study done by someone at DEC about 4 years ago. He tested a SAMBA machine (I don't remember the OS, but I'm fairly certain it wasn't linux) against NT (3.51 was then current, I think), DEC Pathworks, and the AT&T SMB server. The SAMBA machine beat them all hands-down. (Sorry I couldn't come up with a reference, but I read it on the SAMBA-users mailing list a few years ago.)
.c.
Since theyve proved that NT whoops Linux, I guess I'll have to replace my linux installation on my server at home 486-66 (16 megs of ram). And put NT 4.0 with IIS on it so that it will perform faster.. :)
Hee hee
Can you adminster a linux box? BSDI is not that difficult.
James
Look at the OS Configurations:
...
For one, NT used 1Gb of ram will Linux used only 960MB. Surely they could have passed the parameter MEM=1024M to the kernel
Additionally they tuned tcpwindowsize under NT to 65536, and adjusted buffers on the network card to 200 (from 32).
They made no TCP/IP stack adjustments OR adjustments to the netcards under linux.
Just look at the sections explaining the myriad of things they did to "tune" NT. Then look at linux. Enable NFS. The following daemons were run. blah blah. Didn't bother to work on anything.
James
Does anyone remember the Gartner Report about a two months back that pitted *nix based systems versus WindowsNT?
Anyone else happen to remember how bad WindowsNT lossed? In fact, if I remember right, they recommended *cough**cough*FreeBSD*cough**cough* with Apache as the perfect webserver combination.
Lovely progaganda on M$'s part. Go Bill!
www.netcraft.com/Survey/influence.html
shows the most requested sites from the what are they running cgi... reports hotmail as BSD/Apache
Note: Yahoo is freeBSD hosted also...
Widelinks doesn't completely disable symlinks only links out of the designated samba share. So with widelinks=no you can still link within the share. So just testing whether it is a symlink isn't a possibility.
Yep, Samba is nice to beat NT, but guess why they didn't want to test the NFS for file-serving?
Oh, NT cannot do this? *grin*
Shouldn't it be "NT faster than Linux"? I don't agree with the statement, but grammar is neat.
From my meager understanding of web servers, it is not surprising that Apache came in a bit behind IIS when servering static pages. My experience is that for true static content many webservers like IIS and Lotus Go, are faster. Where apache gets major kudos is when running cgi, as the fork and :)
exec of the script environment seems much better. Try running a good hunky perl program on each request and see what you get.
The other thing is that although I am a big fan of linux on powerful machines, its greatest charm is that you can run it on a 486 with 16 megs of ram, and have a relatively well behaved web server that can stay up for 60+ days with no user intervention.
There is no silver bullet. Plus, werewolves make better neighbors than zombies or vampires anyway.
Your comment is essentially: "I haven't bothered to click on the link to read the paper but I wouldn't be surprised to see that Microsoft's cheating." It took me all of five minutes to read the cited paper which carefully listed the exact configuration of all hardware and software used, with justification. You couldn't spend the five minutes to do that but went ahead and spewed your useless comment. Why? -- Dave
More FUD from the FreeBSD advocates. Urban myth anyone?
I'm glad to be in this universe, instead! For one thing, I've never put my "system in a state where you must reload all software from scratch since you need to recompile and reinstall the kernel." Funny, I can recompile and reinstall kernels until my fingers bleed, yet never need to reload all software from scratch. And did you see, that over a two month period, there were six kernel updates?!!? There you have it! I guess that's bad. But, even idiocies like these are just fine, because "MINDCRAFT, INC. SHALL NOT BE LIABLE FOR ERRORS OR OMISSIONS CONTAINED HEREIN, NOR FOR INCIDENTAL OR CONSEQUENTIAL DAMAGES RESULTING FROM THE FURNISHING, PERFORMANCE, OR USE OF THIS MATERIAL." Hey, that's cool. I wish I could not be liable for errors or omissions in my work. But then, I don't work for Microsoft, like they do....
I would gladly flame you, but you seem to base your glowing reports of quality on unreleased software. Is any of that software you mentioned out yet? SQL Server 7.0, maybe? I know the rest isn't. Anyhow, the released versions of those packages blow, and had you claimed exceptional quality for them, you would have earned every flame you got. Although, you were astute enough to say "(at least, from my point of view)." Have you ever used Office 97, NT 4.0/Win95, or IIS 4.0? Doesn't sound like it. Thanks for playing. Have a nice day.
Execpt half the software he named isn't actually released yet.....
Html in notepad.
Do it in emacs.
Yes, the one-all do-all operating system that is worth it's disk space in gold. At least that's the way a lot of slashdotters feel. And one courages soul dare to speak good of the other OS, and a lot of squabbling occurs.
FOOLS. The masses say. ALL FOOLS. Microsoft is evil. Those that use it are morons. ALL morons! 90% of all computer users are morons!
This is hardly rational. The truth is Windows is good enough to a degree that 90% of computer users (some who are, dare I say it, savy) use it, or even, trust it. I know that popularity is not a criteria for OS of choice, not even on slashdot. I remember another small OS that had the same stumbling blocks getting into the open.
I agree with the sentiments of the original poster. Windows deserves credit for what it does well. Doesn't everyone. Do not block an open mind. If you want flame-bait, I'll give you flame-bait! You want a balanced opinion from someone who has USED the software, read the poster. You want bickering, read slashdot. Now there's flamebait.
--
I can't find the url, but I seem to remember an article about how M$ tried to put hotmail on nt to brag about it and then quietly put it back to linux or a unix variation. Even if it was a unix and not linux, that still proves apache/smb is a better combo. And also the url about how ms threatened to leave nasdaq unless nasdaq used nt, so nasdaq used nt on some non-real-time not-very-important task. Anyone remember this?
Stop whining!
This kind of report can only mean one thing.
Doesn't matter how true it is.
We have to check our kernel/samba/apache
code and do it better and better...and then we will show them that even in M$ test, we are better.....
This post should be on slashdot's first page.
Com'on rob!
Well I agree with 'im, and my time is worth plenty to those that fill my bank account. I work on NT. I use Linux for my home network. And my home clients are the most demanding.
I especially agree with the 'dumbest' part.
If the numbers are incorrect, the only thing that's
going to convince me is an actual reproduction of the test,
second best is technical information, which is what
some posters are providing.
Your own contribution ("I've never read a bigger pack of lies")
doesn't tell me anything useful and only takes away credibility
from the criticism posted by others. You are damaging this forum.
Argue the facts, not the circumstance that the report serves Microsoft's interests.
all you linux weenies...if linux is better, take a system that is configured EXACTLY as the box in the test and put the lie to methodcraft.
-- your knees hurt, don't they?
What about stability?!! There is no mention of the fact that NTS has to be brought down periodically or it brings itself down. Or the fact that MS itself acknowledges that it crashes under load. (MS Hotmail runs on Solaris due to this "feature.")
For the cost of two NTS boxes (primary and backup) You can set up several 486 linux boxes to service the same IP and run them all continuously for equal or higher throughput and infinitely more stability.
D. Keith Higgs
CWRU. Kelvin Smith Library
My office has been taken over by iPod people.
Yeah no doubt. The overhead of forking new Apache(s) to handle the large number of incoming connections *will always* slow down a system.
Any good Apache administrator knows this.
Also, under the IIS configuration is says "Server set to maximize throughput for applications when doing WebBench tests", but it *does not* say this under the Apache configuration?!?
Jim
Well, IIS is not much better. Any reason why they had to tweak these settings??
c es: \InetInfo\Parameters\ListenBackLog=200 \InetInfo\Parameters\ObjectCacheTTL=0xFFFFFFFF \InetInfo\Parameters\OpenFileInCache=0x5000 .asp
Used the NIC control panel to set the following for all four NICs:
Coalesce Buffers = 32 (default is 8)
Receive Buffers = 1023
Transmit Control Blocks = 80 (default is 16)
Adaptive Transmit Threshold = on (default is on)
Adaptive Technology = on (default is on)
Adaptive Inter-Frame Spacing = 1 (default is 1)
Map Registers = 64 (default is 64)
SMTP, FTP, MSDTC, and Browser services were disabled
Set registry entries:
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Servi
Using the IIS Manager
Set Logging - "Next Log Time Period" = "When file size reaches 100 MB"
Set performance to "More than 100,000" ? Removed all ISAPI filters
Removed all Home directory application mappings except
Removed permissions for "Application Settings"
Logs on the F: drive (RAID) along with the WebBench data files
Server set to maximize throughput for applications when doing WebBench tests
So much for 'smart/dynamic/self learning'.
Jim
PS - What the hell is a 'More than 100,000" setting?
No one (unless they are fools) sets 'StartServers' equal to 10 and then blasts it with thousands of simultaneous connections.
Might as well put out a burning building with a soda straw.
They should have set 'StartServers' to 500 or 800. Jim
This test just goes to show that every operating system has good parts and bad parts. While Microsoft may be the current lead in speed and usability, Linux is still a little bit ahead in stability (notice I said "a little bit"). I think the main reasons for this are resources and one of the fundamental rules of capitalism.
Microsoft has far more resources than Linux hackers could ever hope to have. While they may not have nearly as many developers working on the product, they do have a relatively closely-organized team all communicating within the group and heading for specific goals. They have entire company departments devoted to testing, user interface design, code beautification, etc. Linux, on the other hand, doesn't have quite as organized a structure. This isn't to say that the development of Linux is worse or less efficient, it's just disadvantaged in some areas, by no fault of its own.
As for that major capitalistic principle, it gives Microsoft both an advantage and a disadvantage. Microsoft's goal is to make money. In order to attain this goal, they must have a superior product in order to get people to buy it (let's leave the monopoly crud out of this for a moment). Therefore, Microsoft strives to make the best, fastest, and most usable product out there. And when they fail, they generally try to fix the problem as soon as possible via service packs, hot fixes, etc. On the other hand, this is also a disadvantage because rather than implementing features that are needed by a few customers, they implement only features that are needed by a majority of customers. This is where Linux has an advantage, obviously, due to its open-source model.
I won't get into it much further than this, but I think we should all at least give Microsoft the credit they deserve. For the most part, they do try to have a good quality product, and their latest work (e.g., SQL Server 7.0, Office 2000, NT 5.0/Win2000, IIS 5.0) has been of exceptional quality (at least, from my point of view). Why not view this as a challenge? I'd be willing to bet that Linux could catch up with and even surpass NT's benchmarks within a year or less.
And yes, I expect flames galore for saying good stuff about Microsoft...
--
Wonko the Sane
And it should be pointed out that my intent was not to defend Microsoft, but rather to try and make people give them credit where they deserve it. I can honestly say that I would prefer to use Microsoft products under many circumstances, mainly because they're easy to use and stable enough for my purposes. Linux on the other hand, is almost impossible for me to use (I'm only a beginner when it comes to Linux, while I've had years of MS experience). I just think Microsoft has a pretty bad and not entirely fair stereotype going against them...
--
Wonko the Sane
--
Wonko the Sane
--
Wonko the Sane
--
Wonko the Sane
--
Wonko the Sane
I think i remember reading a review that
ZD did themselves, that showed what 2 or 3 other
tests showed at 32> clients, Samba on linux is 2.5 times faster than NT's file sharing
The fact that Microsoft won't use their own servers on one of their own sites, shows how much they rely on their product. If an NT server isn't good enough to handle their own web services, why should it be good enough to handle mine?
Chech out this poster on DejaNews Power Search.
Forum: *linux*
Author: will@whistlingfish.net
http://www.dejanews.com/home_ps.shtml
This one got the charts right:
Linux Up Close: Time To Switch
See "RELATED LINKS": The Best Windows File Server: Links & Linux Is The Web Server's Choice
>It takes a little less for me to install Windows >98/NT than Linux. (Of course I lose a lot of >flexibility going MS.)
:)
Just becaus it came to mind, I honestly had a lot less trouble installing a linux distro for the first time then my first win9x install, both times I had someone who knew what was going on with me though, but I knew more about linux than win at the time. I installed nt4 workstation at the same time, a few months later it fucked itself over after about a week of active use, just shit itself then did the old ms popup windows everyfuckingwhere trick so I touched the sky with my middle finger in the direction of bill and sunk the boot in and nt was no more.
Just my 2c
Just noting that Apache runs on NT.
Actually you point out a solid Linux opportunity. Even if the tests were not conducted on a level playing field (and many are not) paying attention to where the bottlenecks were should allow improvements to occur. Nothing would be particularly wrong with an improved Linux regardless as to how it stacks up against other OSes.
:-)
Wonder who would be questioning what if the study indicated Linux was twice as fast as NT?
We run Apache/NT on an old 486 with 48MB of memory on a low volume site on an ISDN connection. It runs very responsively and stable. We attempted to install RH Linux on a twin and ran into issues with some of the older hardware components, so we haven't been able compare in that fashion.
Also, don't forget that the Dell Server was specifically designed with NT in mind. My optiplex also has a little metal sticker on the front attesting to this fact. If you did the test with servers from a Linux vendor, who has picked well supported components and installed a tuned kernel, and the results will be quite different.
Didn't the folks over at zdnet.com do similiar benchmarking (NT versus Linux) with the very same ziff-davis benchmarking tools? And didn't the folks at zdnet.com come up with wildly different benchmarks?
I'd previously put Redhat 5.2 on a Dell 6100 using the same AMI MegaRAID 0.92 driver mentioned in the article. I believe it was using kernel version 2.0.35. Was able to turn SMP and RAID on at the same time with no apparent difficulty, except for the PS/2 mouse not working, which we didn't care about. (BTW, this machine originally ran NT, but we had to switch to Linux to get it to perform it's VPN gateway task effectively.)
It's a bit of a moot point, since we've been buying VA Research machines, but I wanted to know which, if any, RAID driver would be used on a Dell 6300. Now I know the same one that we used on the 6100 works. Thanks, Mindcraft!
BTW, Dell will sell that machine with either NT or Linux preinstalled.. I wonder why they didn't test it that way? Certainly Dell would ship them with all the memory enabled under Linux, and with everything properly partitioned, etc... ?
>PS - What the hell is a 'More than 100,000" setting?
From the Beta IIS4.0 class I took, I recall there is a setting for "performance" where you tell it how many hits the site is expected to get in some time period.. a day, I think. That's the top setting, presumably telling it to dedicate more resources to HTTP serving.
A high profile Linux provider (such as VA Research) should start a Server OS Shootout.
...). At 10:00 the performance test begin. They last system standing wins.
Set up 3 weight classes: single, dual, quad. Publish the specs of the system.
Invite contestants from various consultancies or OS vendors. They arrive at 8 in the morning with their CD in hand. They are presented with a machine in the weight class they are competing in. They can do anything to the system they damn well please other than adding hardware (they may remove parts, but if one of the performance test depends on it, well
The hardware company will get tons of PR. The consultants will surely get loads of work if they present a good standing. MS can show up to make sure NT looks good (ROTFL). But most of all everybody will be assured that the best OS won.
Aah, change is good. -- Rafiki
Yeah, but it ain't easy. -- Simba
The type of hardware that either OS can beat the other on is largely irrelevant. People buy the hardware to run the OS. Note that even Windows 95 blows Linux away on PPP performance on WinModems. A fair test would be to take two similarly-skilled engineers (one an NT guru, one a Linux guru), and give them hardware budgets to be spent at one soup-to-nuts parts vendor. Heck, spot NT the licensing fees. Put in the sort of requirements one would have for a real-world server (such as hard drive failover), so that you have to tune for stability and speed. Which OS can do better on a Quad? Frankly, who cares? As was mentioned upthread, the extra three processors probably slowed Linux down for this test. Which OS can do better on (say) $5,000 in hardware? That's the important question to my mind.
--The basis of all love is respect
You have to edit a few lines in the kernel to get it to support more than 960 MB of physical memory.
What I found interesting was that they apparently didn't make a separate swap partition for the linux box (they said 1 OS partition and 1 data partition)... hm...
A while back, dell.com ran on a single Sun box - I think it was a 2-way Ultra II running the Netscape server. Microsoft noticed this and approached Dell to see about changing this over to an NT-based system. Dell hesitated (things were working well) until Microsoft came up with several million dollars as incentive. (They hardly needed to do that, since all they had to do was even hint that Dell might not get "most favored nation" pricing, but it shows how badly MS wanted it to happen, and quickly, so they funded the switchover.)
The lone Sun (actually there were two for redundancy, but a single one was capable of handling the total max load for what was at that time the world's busiest e-commerce site) was replaced by OVER A DOZEN (14 if I recall) 4-way NT servers.
Ask me how I know - Before deciding to leave Dell, one of the jobs I considered and interviewed for was heading up their internal Sun computing. (Don't believe for a second that they really "run Dell on Dell" the way Sun "runs Sun on Sun". So far as I know, Sun and IBM are the only computer companies on the planet capable of running entirely on their own gear. Dell will not be able to replace Tandem and Sun for a very long time yet.)
To construct a suspect environment where NT/IIS looks good and Linux/Apache looks bad is easy
(as is finding analyst shills, unfortunately), but savvy buyers will do the math and find that real-world configs of NT/IIS will cost far more money than capacity and capability equivalent Unix-based alternatives.
FUD continues to spew forth from Redmond... Are we really surprised?
"The future's good and the present is nothing to sneeze at." - Roblimo's last
If you look at the certification section it is "mindcraft" admits that MS paid them to do this benchmark. Frankly, I can't see how that could possibly make an objective benchmark.
Why shouldn't it make an objective benchmark?
Coming soon - pyrogyra
The thing is... why doesn't Samba test symbolic links in a performance friendly way? It would appear to be brute-forcing it, which seems pretty poor to me.
Surely there's a low-cost way to check whether a file is a symbolic link or not? Or even do a pass over the filesystem occasionally and check the links that way. Or just cache as it crops up.
You should set widelinks = no. However, doing that shouldn't affect performance that badly.
Needs fixing.
Coming soon - pyrogyra
Arbitrary cutoff? So that it doesn't spend forever chasing down links?
Coming soon - pyrogyra
Well... here's the docs on boot.ini:
o d98/html/3952/table1.html
http://msdn.microsoft.com/library/periodic/peri
As for physically yanking the RAM out of the box... dunno. Maybe it was less hassle to use the switch.
Coming soon - pyrogyra
See what I mean? It used 1024Mb.
Coming soon - pyrogyra
From Andrew Trigell (original Author of Samba):
:)
They set "widelinks = no" now I wonder why they did that
My guess would be so that you would have a secure system, which is what 99% of admins not trying to rig benchmark results would arguably prefer.
widelinks = yes gets you hit on the nose with a soggy newspaper, if you're an admin.
Coming soon - pyrogyra
The first thing I did was scroll down to the end to see who sponsored the thing.... Wasn't surprised at the result! After that I read it for humor more than anything else. Look at Samba's website for some other performance tests against M$ NT$, I'd like to see them compare NT running on a Sun or HP box! I like the comments that it requires technical knowledge to setup Linux/Samba/Apache - typical stupid comments! Ahem, so let me get this straight - let's make the network very easy to install so the MC$E that conned the manager in hiring them look good. Now when there is actual problems, this MC$E doesn't have the technical brain to figure out how to fix it! When are people going to learn it's not the initial purchase and installation of a network that's expensive, but the daily maintenance, troublshooting, especially downtime is. I for one thank M$ for the constant onslought of shitty OSes, I know I will have a job for life! Thanks Linus for REQUIRING anyone who uses Linux to have the technical knowledge to understand it!
Wow! It seems like all of our true anonymouse cowards have come out of there holes.
I must say I find it amazing how quick a lot of our fellow slashdotr's are to judge. I work on a production team for a major high tech company. A lot of our Database servers run NT, Anything that is doing security is running some flavor of Unix or Linux.
Just today my mentor(Yes im an Intern, and 16 at that) was telling me we may have to add 3 BSDI servers to our network. I was floored! We already have about 5 diffrent OS's on our network(thats counting diffrent flovors/versions of unix and Linux)
When i asked her why the answer was pretty simple. The program out grand high mucky mucks had dessignated for Ecommerce was optimized for use with BSDI. Now i have to learn yet another OS.
I enjoy hte learning, its why im here, but I have learned one thing. SQL doesnt work well Linux boxes. NT4 has a few security holes that NT5 doesnt. You can fix those with SP4 but then our back end doesnt work.
Ive learned how to set up security on a Unix box in such away that it will make the NT boxes more secure.
I have also learned that most OS are really good at a specific task, and that Zealots will use that to the best of there advantage.
Every OS i have used had some really positive points and some really good points. Linux is great, But i wouldnt want my grandmother using it. The MacOS is great when you want simplicity or are doing major graphics. Windows on the other hand is great if you want to do just day to day stuff. I would hate writting a term paper in Vi or Emacs. Heck i would even hate doing it in Pico! But Word works really well, and notepad can be the "quick and dirty" programmers best friend, especially with HTML stuff that you need NOW.
Everything has a positive, everything has a negative. Lets not get to angry at those who are willing to actually admit they agree with somthing we dont. If everyone was like that this whole world would be like Kosovo.
Lissell (where have all the cowboys gone?)
I'm sure that you've heard how fast photoshop is on the mac, well that is due in part to the fact that the apple long ago put in special calls that photoshop uses. These calls don't exist on the pc side however. The powerpc chip itself is faster clock for clock than anything on the x86 side, however the boxes created from it suffer from an obsolete operating system and poor motherboard design. Take a powerpc chip and put it on a good motherboard and put something like Linux or BeOS on it and you've got a very impressive machine. These aren't the machines that apple is selling however.
Muslim community leaders warn of backlash from tomorrow morning's terrorist attack.
Linux as we all know runs very well on single processor machines. It also runs well on 2 processor SMP systems. But above that the principle of diminishing marginal returns begins to kick in hard. NT simply makes better use of resources on SMP systems than linux currently does. This is not suprising since few linux devlopers own a 4-way SMP system whereas Microsoft can buy all they could ever want. This is a rigged study. Apple pulls similar tricks when they compare their G3 systems to x86 machines. If Mindcraft were to compare Linux and NT on a single or dual processor machine I'm sure the results would be quite different. As SMP systems become more affordable and commonplace I'm sure that linux will catch up and likely surpass NT in this area. Microsoft claims that NT will scale up to 32 processors but the truth is that it begins to decay rapidly above 4 processors and so the version which is supposed to work with more than 4 processors is rare and not commonly available. Mindcraft would do well to throw Solaris into this comparison, but they won't do that because NT would get it's clock cleaned.
Muslim community leaders warn of backlash from tomorrow morning's terrorist attack.
Any benchmark or test should be validated before being taken as gospel. Be that as it may, and assuming the benchmark would survive scrutiny, several important factors are missing from this performance survey. First, NT performance and stability degrade consistently over time, eventually, and in every practical case I have seen, resulting in total system meltdown requiring reinstall. This is an especially important factor in filesystems, where NT shows significant degradation on the order of weeks. Next, the benchmark ignores the possibility of using Linux as an NFS file server, versus running an NT based NFS server. NFS is a significantly lighter protocol, not to mention the fact that there are NO decent NFS servers for NT. Moreover, the underlying NT filesystem can not support most of the permissions and case requirements to serve UNIX files, effectively.
Maybe Linux isn't a perfect surrogate for NT, but it is without a doubt a better solution overall.
John
Microsoft admitted during the Antirtrust trial that they actually tell them(people conducting the surveys and test) what results they want.
Wonder about mindcraft dealings with Microsoft? A quick look at mindcraft's main page makes you wonder even harder.
... what a surprise, they did some good efforts there as well to prove NT is much superior to each of the other alternatives (NT was 2.6x faster ...).
...
In addition to the Linux vs. NT, there are two other report: One that compares WinNT to Novell Netware 5.0, and another and compares WinNT to Solaris on an UltraSparc, and
Now, you may wonder: who sponsored both of the other tests? The NT vs. NetWare was sponsored by MS, and I bet the other report was sponsored by just the same company.
It is interesting to note Mindcraft didn't expose at first who sponsored their NT vs. Netware study, as the client "did not authorize us to disclose their name". Only after publicly critisized by Novell, they published MS payed them for an "independent and unbiased assessment" - yeh, sure.
And if you still need more evidence about Mindcraft dealing with Microsoft, MS also payed them to prove that microsoft did nothing malicious to cause Quicktime plugins to fail. This was, if you remember, they main claim of apple against MS.
It's always good to have friend$$$ in the business
The benchmark reflects my experience pretty accurately.
Some Caveats
1.) My vendor is Data General and DG has offically said that they do not support Linux in any way shape or form
2.) The hardware I'm running on is Intel based, but based on #1, getting Linux to even boot successfully is difficult.
Basically we have DG 3700's and 3800's (Basically 4-way Pentium Pro units, and 4 and 8 way Xeon units). The performance on them in testing is consistent with they described. BUT remember this key detail. How many Linux developers have access to type type of hardware? Espeically if the vendor has said they don't want Linux support.
Additionally, Dell has been somewhat anal at times about releasing specifications. But that server the 6300 is pretty common, I'm sure someone has access to a similiarly equipped machine that they can use to recreate the results.
Any takers? I've always wanted to find another outfit running DG hardware under Linux but I don't think it exists.
The benchmark reflects my experience pretty accurately.
Some Caveats
1.) My vendor is Data General and DG has offically said that they do not support Linux in any way shape or form
2.) The hardware I'm running on is Intel based, but based on #1, getting Linux to even boot successfully is difficult.
Basically we have DG 3700's and 3800's (Basically 4-way Pentium Pro units, and 4 and 8 way Xeon units). The performance on them in testing is consistent with they described. BUT remember this key detail. How many Linux developers have access to type type of hardware? Espeically if the vendor has said they don't want Linux support.
Additionally, Dell has been somewhat anal at times about releasing specifications. But that server the 6300 is pretty common, I'm sure someone has access to a similiarly equipped machine that they can use to recreate the results.
Any takers? I've always wanted to find another outfit running DG hardware under Linux but I don't think it exists.
The statements that I will make are not biased. :)
I agree that Microsoft products are a little easier to setup. It takes a little less for me to install Windows 98/NT than Linux. (Of course I lose a lot of flexibility going MS.)
If so many Linux users claim that it is faster and more stable, then there must be _some_ kind of truth to that. Sure every OS has pluses and minuses and that MS has more resources because they are rich. MS might strive to make their product as good as possible, but there is a great deal of improvement. However, Because of the beuraucracy of such a large company, those service packs take a long time to be released. Look at the average time between kernel releases and service pack releases. Windows 98's service back was due to be released in December, but it was pushed back. Meanwhile, I got several new kernel sources. Linux is not only the kernel, but the additional software that actually does the networking, and serving, etc... Those have been updated countless times since, also.
As for my personal experiences with Windows 95/98/NT4.0:
They need to be reinstalled at least twice a year due to corruption. (Particularly in the registry. I'll post about that when the time comes.)
The system itself is visibly slower than Linux on some occasions.
Starting and stopping some services require a lot of waiting time.
'98 crashes the most, NT is middle, Linux not once (unless I run Wine
The reason I switched from NT4.0 to Linux was because I got tired from the above, and I have found sanctuary with Linux.
I guess I can keep going on about my Windows problems, but I'll save it for later.
TeslaCoil
at least that is what it was before they changed it, I think it's still something similar
The real test doesn't come from who sponsors the test but if it can be reproduced. I think another team needs to try the test again with the same configuration and see if they get the same results.
We, as a community, should also be on the lookout for any configuration option that could have been changed to make Linux work faster. It could be that they wanted to ruin Linux with a crappy config.
Don't be so fast to jump.
MINDCRAFT, INC. SHALL NOT BE LIABLE FOR ERRORS OR OMISSIONS CONTAINED HEREIN, NOR FOR INCIDENTAL OR CONSEQUENTIAL DAMAGES RESULTING FROM THE FURNISHING, PERFORMANCE, OR USE OF THIS MATERIAL.
I don't know if you did this in high school or college, when you had a result that you knew couldn't be right, you forged the data within 12 percent of what you expected, and then at the end of the paper, you put in your conclusion, what could be the cause of such errors. I think they forged the data to what they wanted, because another test done by Micrograf proves differently. What I would suggest is Redhat or someone else proform the exact test and see what yah really get. I know there was a paper a while back about 2 years ago that showed the defects in NT's 4.0 proformace. the link is still around http://www.ugraf.com/unix-nt/jt/
Check it out, pretty interesting.
I don't know what life is, but no one gets out alive...N
Hey guys I know I had written earlier but I came across another story about Bleeding edge doing their own tests.
. shtml
here is the results:
http://www.gcs.bc.ca/bem/editorials/nts4rhlinux
I think there will be more test to come proving once again Microsoft has fixed the races, not just the modem *smirks*
I don't know what life is, but no one gets out alive...N
Well, to set the record straight Novell has decided to do the bench mark, which is better than redhat doing it themselves. Hehehe, but you got to admit the previous story was funny.
I don't know what life is, but no one gets out alive...N
This whole issue brought about a new question for me:
When they set up this Linux machine....did they actually have someone who knew how to set up a Linux box. If you take Linus Torvalds Redhat 5.2 (I'd say he knows how to set up a Linux machine) box versus any Microsoft Professionals' setup of an NT station...I would like to see that test.
Just my thoughts on the next time they test performance
I work for a consulting/ISP company. We run both NT 4.0 w/IIS 4.0 and Linux (RH 5.2) w/ Apache. We run comercial websites off of our NT and Linux boxes with our user pagers are only served up by apache. We run 2 nt boxes to be able to run all the NT APPs (IIS, exchange, SMB, SQL etc..)and still have reliable perfomance for our internal network and for our customers. Both NT boxes are on dual 266 pentiumII with 128 megs o' ram and we still have to devide up services between the boxes so they're not bogged down all the time (tried running everything on one box and it sucked a$$). Whereas our linux box is a pentium 75 128 megs o'ram and serves up http, smtp, dns, pop3, samba, routing and a whole slew of other services. In short to be able to run all the services we have on pentium 75 linux box (very very cheap)would take several dual PII266 boxes (at about $1600 a pop). And I won't even go into how long each machine says it has been up for. So there is MY bench test the old tried and true real world bench test.
root@localbrain root>ps ax |grep thoughtd
I just started working with linux about 1 year and half ago and when I started working with linux the first thing I did was load it on my old 486 at home. I bought a book and was able to install everything with a minimal amount problems (took me like a whole saturday). It took me about 3 days to tweak everything to get my box to run at optimal levels, and not at any point did I find it difficult to find help or documentation. I went from knowing absolutely nothing about any flavor of UNIX to serving up web pages and having my own mail server in less then a week, and my box was screaming. It is was all due to the fact that I could get good documentation and I could find help. Posting a few things to a news group doesn't count as looking for documentation. If they had a problem getting info then they just didn't try very hard. Of course this is based on the assumption that these people are farily intelligent. If they are just morons then yeah I can see the problem.
root@localbrain root>ps ax |grep thoughtd
Linux can run on a number high powered RISC machines. Another common solution for Linux is clustering. These may be a better solution for high powered Linux requirements.
They claim that reinstalling the Linux Kernel requires all software to be reinstalled. Assuming that the tests were entirely above board, with this level of understanding they would still be unable to do any real fiddling with Linux, however they seemed to know a lot about NT.
They seemed to do a bit of whinging themselves. It is ridulous that a couple of questions posted to a Linux newsgroup should be expected to compete with technicians who live & breath NT.
It would be good to see this expirement verified. Possibly it is valid, but Linux or *BSD would be likely to win if it were done on the basis of $ for speed comparison, or if it were done on more Linux friendly Linux platform.
Other reports seemed to show a significant lead for Linux, even excluding Licences. It would be nice to know what the difference was.
We use GNU/SunOS.
Something was obviously very wrong with their benchmark. I wonder if they did something funny, like disabling SMP for linux? Convenient mistake when compiling the kernel?
...These aren't the droids you're looking for....Move along....
Microsoft the credit they deserve? You go to www.netcraft.com and see how many servers out there actually run Microsoft IIS. You will find far many more running Netscape/Enterprise and Apache. Microsoft's own famed Hotmail runs on unix! When they tried to move it from Unix to NT it CRASHED HORRIBLY.
NT can't handle the big loads. On smaller loads it is known to show more performance, but you start pounding that sucker and it goes to hell. Unix and Linux keep on chuggin.
Did you not see the emphasis that this report put on the lack of support for Linux? Did you see how they fine tuned the nics and the CPUs under NT and not on Linux? Microsoft is trying to cause FUD in the minds of everyone.
The only respect they deserve is that they have one hell of a marketing scheme. Oh yeah, Bill Gates is good at legally pirating software and calling it his own.
Madhatter --It's no wonderland out there.
Sorry for the flaming earlier. I'm ready to go home and a garbage NT machine is keepin me at work. But hey, the PC is only as good as the operator.
Madhatter --It's no wonderland out there.
Just look at where Linux and Apache have made it to in the real world. The tests they conducted were so far fetched it is really hard to believe. There are alot more big companies out there that use apache and Netscape/Enterprise. IIS just can't handle the traffic that they can. Microsoft's Hotmail dosen't even use IIS because it handle the load of nearly 1 million users.
The only thing Microsoft has for them is that they are easy to install. That is what they are trying to capitalize on here, and it looks like its going to work very well even with what the linux news web sites will have to say. FUD all the way.
Madhatter --It's no wonderland out there.
No kidding man. The dude is a total NT deadhead just following the instructions microsoft gave them. They did not follow an optimal Linux config in any way.
Madhatter --It's no wonderland out there.
>>Apple pulls similar tricks when they compare their G3 systems to x86 machines.
, 4161,2232618,00.html
e ed.html
**Excuse me?** Would you like to provide facts to back this up?
Even PC Magazine, a company that almost no one would argue would be biased in favor of Apple vs. Wintel, admits that a G3/400 beats a Dell Dimension XPS T500 Pentium III on tests using an application *widely reported* to be heavily optimized for MMX and SSE! All that with a 20% slower clock speed!
http://www.zdnet.com/products/stories/reviews/0
Perhaps you've heard of Project Appleseed? No?
http://exodus.physics.ucla.edu/appleseed/apples
Imagine that--an independent study of G3 vs. x86 and other processors...hmm...a PII/300 is over 33% slower than a G3/266 (13% slower clock), and over 17% slower than a Rev. A iMac (29% slower clock).
Not to mention that a PII/450 is 8% slower than a G3/333 (a clock difference of over 35%)!
Keep your uninformed crap to yourself. The discussions of G3 vs. Pentium II/III proved long ago the superiority of the PowerPC architecture.
Unfortunately, tests like WebBench and NetBench hava a _lot_ of subjectivity built in. For example, the WebBench homepage WebBench states "WebBench also has dynamic test suites where we've re-written the CGI application as:"..."a Windows NT Microsoft Internet Server API dynamic link library". With that in mind, MindCraft could have easily run the native API dynamic tests on NT, versus the "regular" tests on Linux, greatly biasing the results. Yet their "white paper" never becomes specific on exactly which tests were performed, only the overall results. Hardly conclusive.
--
"Modern cryptography is nothing more than a mathematical framework
for debating the implications of various paranoid delusions".
Thanks! I was wondering where (or if) this important piece of data was in their paper. I do, however, notice several discrepancies in the data: . The pagefile for NT was set to 1024MB, but no mention is made to the Linux swapfile. . The claim was that Samba 2.0.3 was used, but the config spec indicates the version to be 2.0.1 . No mention was made regarding compiler optimizations made to Samba during compile. Furthermore, (and this may be a personal bias, but what the heck) this article is completely subjective. For example, there is a section titled "Observations" that attempts to make statements about issues encountered during the testing. Where, then, is the section for NT and IIS? Surely there were some issues there. Instead, we see the same Microsoft anti-Linux pitch all over again (codebase in flux, chaotic documentation and support, etc). This is hardly objective.
--
"Modern cryptography is nothing more than a mathematical framework
for debating the implications of various paranoid delusions".
Microsoft Corporation sponsored the testing reported herein.
When Bill Gates wanted 85% of IE uses/developers to favor OS integration, a Microsoft sponsored a survey which produced this result. After further investigation by the court, it was found that MS PR felt the questions were very leading and invalidated the results.
When Bill Gates wanted to show how "easy" it was to install Netscape, a testing lab was commissioned to test and video tape the results which showed it took just minutes to install Netscape. After further investigation by the court, it was shown that the film was doctored to eliminate a large, time consuming step.
When Bill Gates wanted to show that an IE removal program slowed down Windows 95/98 and broke windowsupdate.microsoft.com, a film was produced showing exactly that. After further investigation by the court, it was shown that this film was doctored as well and actual parts of the test showing poor performance were actually from machines WITH IE installed. When MS tried to replicate these "authentic" tests, they could not reproduce any system slowdown after IE was uninstalled.
Microsoft's record speaks for itself. Their testing credibility is worse than OJ Simpson holding a bloody steak knife.
I didn't notice one word about whether or not they had X-Windows up or not. And if they did, what window manager were they using, what programs were running, etc, etc.
This "study" is about as valid as the tobacco industries on the health risks of cigarettes.
STFU & GBTW
I have installed just about a billion NT server and can do it backwards forwards sideways and under. 95 percent of my installs go with no problems. However NT is neither easy nor intuitive to get to perform properly. And after all my experience it still took me 2 days to get my dad's NT 4.0 with a built in version of SP2 to work correctly. Due to a but with SP2 and RAS, ended up having to hack the hell out of it.
Still I have had my problems with Red Hat 5.0 and the never quite correctly recompiling kernal. Slackware didn't have the same problems. But that is besides the point. Every OS has it's purpose. I love my NT box for certain things, my Mac is good for other stuff, and I love Linux's flexability and ability to morph into usefull things (firewall anyone? router?).
In the end, every OS is best in it's designated environment. Linux ROCKS for web stuff. IIs can stuff it!
My 2 cents
-Kit
Read the report. There seems to be an issue that the people whom did the benchmark didnt/dont know how to tune Linux, Samba & Apache. However they did know enough to disable NT services and edit the Registry (which MS states is at your own risk), and dedicate a CPU per NIC. Multi-threaded ? Multi-tasking ? O/S.
There are enough comments from knowledgeable people posted to this article to indicate that the knowledge to tune Linux to outperform NT does exist.
Here is how to fix the problem. Offer to support Mindcraft to tune Samba correctly, recompile the kernel to use 1GB of RAM and fix the Apache configuration. Then challenge Mindcraft to post the result of the re-run benchmark. If Mindcraft wish to be seen as accurate and un-biased they will jump at the chance to re-run the benchmark with a correctly tuned Linux server.
Otherwise simply sounding off in a forum like this only serves MS's purposes further, by allowing them to get your focus on irrelevant activities, not saving the world.
P.S. I too expect Linux, Samba, Apache properly tuned will outperform NT.
Yes, I gather this from what has been said before.
However, Linux kernel can be patched, it's one of
the main strengths of Linux. It reminds me of the people saying "There is no SMB support under Linux" because you have to recompile the kernel to make it work !
Check this page out:f tcheck.html
;-)
http://www.novell.com/advantage/nw5/nw5-mindcra
Starts out >>>
Reality Check
You can Fool Some of
the People Some of the
Time....
Mindcraft's "A File Server Comparison:
Microsoft Windows NT Server 4.0 and
Novell NetWare 5"
For the second time, Mindcraft, Inc. has published a
product comparison of dubious origin, unprofessional
methods and biased results. Novell is embarrassed to
respond to such prejudiced work, but does have an
obligation to inform the market of the facts and publish its
own benchmarks with accurate and reproducible results.
Lots more in the study.
FWIW a search of Novell's site for Mindcraft" turns up some studies that are apparent;y pro-Novell, so Novell is not ashamed to quote 'em when they're good
"A gun is a tool, Marian. No better, no worse than any other tool. An axe, a shovel, or anything." Shane (1953)
http://www.zdnet.com/sr/stories/issue/0,4537,39802 2,00.html
, 387506,00.html
has the comment: ``We know that Linux beats NT in this arena (see "Reviews: Linux Takes On NT,")'', which links (after several jumps) to:
http://www.zdnet.com/sr/stories/infopack/0,5483
which starts out with:
January 25, 1999
Linux Up Close: Time To Switch
Commercial Linux releases are ready for your server business.
By Steven J. Vaughan-Nichols & Eric Carr, Sm@rt Reseller
Forget Linux's hype. Forget Microsoft Corp.'s
server market share. The bottom line, according to
our hands-on analysis, is that commercial Linux
releases can do much more with far less than
Windows NT Server can.
"A gun is a tool, Marian. No better, no worse than any other tool. An axe, a shovel, or anything." Shane (1953)
More proof:
:() has a nice chart that exactly counters Mindcraft's charts
9 6115,00.html
.......
This article from ZDNet (couldn't find it last night
http://www.zdnet.com/sr/stories/issue/0,4537,21
January 25, 1999 Issue
Linux Is The Web Server's Choice
By Steven J. Vaughan-Nichols & Eric Carr, Sm@rt Reseller
WebBench measures the performance of Web
servers in responses per second.
The answer: Linux with Apache beats NT 4.0 with
IIS, hands down. SuSE, the least effective Linux,
is 16 percent faster than IIS, and Caldera, the
leader, is 50 percent faster.
(more text at the site)
"A gun is a tool, Marian. No better, no worse than any other tool. An axe, a shovel, or anything." Shane (1953)
Can anyone cite a reference regarding what OS HotMail is using? I'm not disputing these assertions, its just the first I've heard of it. I've had nothing but problems using HotMail, so I wouldn't be surprised if they made a change.
Why not get a real server, rather than some intel garbage? Like a Sparc UE or something?
Even if it is a quad xeon 400, saying that NT posesses 'superior scalability' is like saying is has 'mature engineering' rather than code bloat...
---------
Get back to me when my brain starts working.
I'll qoute good ol' Winston Churchill on this one.
"There are three kinds of lies. Lies, damned lies, and statistics."
In the land of the blind, the man with one eye is king, and the man with the computer is pretty much ignored.
By latest 2.0.3 version of apache, are they implying apache 2.0.3? I thought the highest was 1.3.6?
http://www.unix-vs-nt.org/ is the site that talks about M$ moving to NT, then having to go back to solaris.
I say, we must start doing these tests.
What about VA Research. They have got the hardware.
What about Comdex? Maybe someone could set up a Linux file server and ask Microsoft to do the same?
And test these beasts under serious presence from VARS IT's etc...
What about scheduling such an event?
What about telling the press about it, once the people who would do this contact microsoft to check that test in public, with confguration and all!
They bring their hardware, we bring ours, same type and all, they configure theirs, we configure ours. They must also have system loads etc...
If they shut a service we shut the same service.
Why not do this and shame them! They might say no, but the press would know about it!
Let's act about this at COMDEX Spring!
If anyone knows SAMBA people and VaResearch people and Linux et al., propose this to them!
Kill Microsoft? No! Just hire their GUI guys!
I agree their methods and configuration must be questioned. In defense of Linux this month's Microsoft Certified Professional Magazine has a focus on Linux/Samba and NT (Front Cover). The person who wrote the artice works for SGI and is on the Samba Programming Team - All this in a PRO-Microsoft Mag. He even states that in their tests that NT 4.0 is faster when serving UP 16 Client requests but under HIGHER LOADS Samba/Linux performs better. THIS WAS IN BLACK and WHITE.
...Only on a lightly loaded server, with 1 or 16 test systems, does Linux/Samba outperform Windows NT Server and then by only 26%....
Another thing about Mindcraft's testing was
This is contrary to Statements made in the May MCP Magizine. I find that they both have their strengths and as a MCSE have seen NT systems that are properly configured perform their server duties well (with careful observation and maintenance). I have also seen Solaris/Linux/AIX all perform as well as or better in the same environments as NT.
Just 2 Cents from yet ANOTHER MCSE & LINUX User (Bet you don't see that everyday...)
Like a Pentium 133 w/ 16 MB RAM? I'm not even sure if NT can run on that at all, but Linux should run pretty well on that config! (I'm not sure though how fine it will run with all those web hits and file requests)
Whoever did that test obviously didn't take full advantage of Linux's open source aspect and doesn't know C. Who in thier right mind would believe that gibberish crap.
MS marketing defies reality again. Any MS product
can be installed, getting them working and keeping them working is not easy. After 10 years of using
lots of different stuff on MVS/Unix/MS, I find keeping a predominantly MS site ( NT/IIS/Proxy/Exhange/SQL-Server/VB/VC++ ) on-line a pain in the ass due to such things as
upgrades that screw up other products, DLL hell etc. etc.
actually.. i was under the impression that hotmail was running bsd. either way, it still proves a point. if the dork really needs evidence, he should go to microsoft.com and try to find something using their pimp-daddy search engines..
the damn thing crashes half the time.
jharrol@comp.uark.edu
-- john
could you elaborate on this. i am not familuar with the network code.
you can email it to me if you want.
jharrol@comp.uark.edu
-- john
If you piled on the ram (so the NT kernel don't gobble up all the memory, mines running at 22 megabytes), disabled all the unneccasary services (if you have protected storage enabled, expect NT to slow down considerably), used a multithreaded server other than IIS (such as Xitami) and ran without the GUI, I could see NT being faster than Linux. That is just not gonna happen. All microsoft products are designed big and slow, but than they develop technologies to accelerate the machine (such as ISAPI). Geez, if NT wasn't so SLOW, they wouldn't have to accelarate everything. Right now I am browsing the web, wondering in the meantime, why the kernel + IE4 need 61476 bytes of memory. And if I start word, I might as well go for a coffee. Why does Word need 10 megabytes of memory without any open documents, and why does it use 90% of the CPU time when it isn't doing anything?
I meant 64176 kilobytes
I don't know much about Linux, but I do know how to make NT perform like a dog. Turn it on. LOL!!!
You don't have to debunk this marketing move. Linux and Microsoft are not in the same realm, and they never will be. Windows is for computer illiterate people who want to "point and click" and Linux is for power users. Most people I know that have windoze machines don't know a thing about them. I know tons about windows 95, 98 and NT just from fixing the computers of friends and family. I don't like it, and as far as I am concerned, people who don't know what they are doing should not be using a computer. BUT, windows has become a GAME machine, and an effective way of creating/dodging work. Get a program that does the work for you, and you don't need to know what you are doing. A HUGE portion of windows software falls in this category. The downside? The user doesn't know what he is doing, so he is always breaking his system. In the end, because the user doesn't know what he is doing, he and his system become a liability. For linux to garner "mainstream acceptance", it would have to become feasible for the computer illiterate. This will never happen, because they can break anything. As for the user who picks an NT box over a Linux box as a server, probably did not get the advice of someone who knew what they were talking about, or did not listen to them. And the local computer geek doesn't qualify, (unless they have some actual training)
Hmm. I installed the apache sources and took a look at the interesting things and I found the following in conf/highperformance.conf-dist: MaxRequestsPerChild 10000000
So, if ( ! IF ! ) NT is 3.7 times faster (max) than a complete solution for Win NT + IIS (for 150 users) must be sold at 3.7 x RedHat 5.2 Distr. (or less to be competitive). Good luck, micro$oft!
Anyone with a small amount of knowledge of Apache
will realise that setting MinSpareServer settings
to 1 will result in *seriously* bad performance.
And Mindcraft claim that "both servers were
optimised".
I wonder how much Microsoft bought Mindcraft's reputation for? Was it worth it?
1. Microsoft commisioned the report, and probably did not want any price performance information in here.
2. 3700 hits per second is probably equivalent to 10,000 intranet users making 20 hits per minute.
3. The cost of client access licences CAL's for a 10,000 intranet would be pretty large (read x00,000 of dollars)
4. The cost of client access licenses to generate that load on the SAMBA server would also be very large.
5. Strange that I recieved a notification of this report from a list that I do not remeber every having subscribed to. I wonder if MS commisioned someone to spam out this report, hmmm. Microsoft sending spam mail, they must be VERY worried about Linux.
That is it.
With all the money you have, Bill, surely you could hire more professional spinmeisters.
Hey, I think Stephanopolis is available!
If not, send me the money. I'll write a glowing report. BTW, there's a substantial extra charge if I have to use the words "stable" or "dependable" anywhere in my whitepaper.
Irv
Well, trooth is, NFS+anything really sucks. There are evolving technoligies that will outperform said platform. There was (is/still) talk that Samba is better...
Any comments on Coda? Andrew?(even tho this may be dead)
to me this seems the most reasonable reaction, how
come nobody accepts the challenge?
There seem to be several MCSE or comparable
... ?
qualified persion in the crew that made that test.
The tuning steps they took were not that obvious
than the average manager would think.
(Having such a person hired is indeed costly.)
On the other hand their report suggested that
the linux part was done without any knowledge
of the system. Having to read the HOWTOs dosnt
proove of much expirience. If they had hired
a Linux professional as well then the situation
were much more balanced.
If i call the *desktop* hotline of MS then i
will get a info compareable to a calling at RH.
But if i pay for premium support (and i will do
this if i have to do business of that server size)
then will get lots more, sure.
I browsed the other slashdot postings and found
some of them very interesting in technical aspects.
I am sure lots of that info will go into an
enterprise-setup-HOWTO or similar, some other
things will merge into applications or kernel.
This already works against that (pretended)
non-knowledge of the testers and benefits
everyone in the linux community as well.
Just one is clear to me, this sort of FUD
approach can never ever happen again. The
sponsor has caused a counter movement with
lots more power behind it. I am prettey sure,
the shootout will and has to be repeated at
honest conditions. And the winner is
Well i dont know totally for sure. But the
price Linux has to win is a price of honour.
Get the machines, get some expirienced system
administrators and choose a place to happen.
Make it in a public environment and the result
will proove the performance, the reliability
and the integrity of each candidate.
Bye AlexS.
PS: Anyone that knows what exact effect of the
different stripe size on the RAID0 device
can be? Is it true that Linux is currently
limited in that area? Or is it another FUD?
--- Linux has no limit !!!
It's very easily to say "Oh, this was sponsored by Microsoft, so I just discard it". And no doubt, a gazillion posts here in slashdot will say something like "it was a bullshit test" and "no matter what they say, Linux is faster". But repeating a mantra doesn't make it true. And it certainly won't convince anyone who has to decide what to use.
If you think they are wrong, prove them wrong. They told you what tests they were using, they told you the configuration parameters. Repeat the test, and publish your results. If you think the kernel can be better tweaked, by all means publish results that are better. That will help the Linux community infinitely more than discarding the results because they were sponsored by Microsoft.
Before you argue "the Linux community doesn't have the money to do such a large test", let me point out that such an argument works in favour of Microsoft. If Linux can't do such a large test, but Microsoft can, wouldn't it be more logical that Microsoft does perform better under heavy load, because they have the resources to test it, while Linux doesn't?
I would like to see Linux perform better than NT, perferably in a standard, repeatable test.
--- Abigail
Well, maybe they did. But that's easy to find out, isn't? You know what test they did, so you can repeat it. This time, you enable SMP and you publish your results.
--- Abigail
What keeps you from falsifying the tests?
--- Abigail
Certainly. There's a lot to be improved on Perls performance.
--- Abigail
Insane.
Diehard
Please submit any inconsistancies you see in this document (and if you don't see any, please shoot yourself in the head) to lwn@lwn.net. They are readying a response as we speak.
Of course NT outperformed Linux. It had FOUR TIMES THE BLOODY MEMORY TO WORK WITH!
Sheesh.
Identical hardware, my ass.
- Pizza
-- I ain't broke, but I'm badly bent.
Don't know why this is moderated so high. Flame bait or no, the fact that you blindly accept the false claims of this oh-so-unbiased "benchmark" should preclude this from a 3 rating.
I wonder why they used such a high TCP window!!
I got a theory...
The window should change dynamically but AFAIK this is implemented in a terrible way in W95. If it is the same in NT4.0 the window will be almost static!
I think this test machine sends out packets
as fast as it can without waiting any acknowledge.
This requires a lot of buffer space (a nice way to waste 1 Gig of RAM) and boosts the figures for this particular scenario (one way traffic).
Not actually a real-world situation, in my opinion...
....that they dont know how to properly set up the Linux box.
I seem to remember a certian Walnut Creek FTP server running FreeBSD on a single CPU in a single box holding the world's record for the most bandwidth used in a single day.
Didnt it take 27 NT boxes to equal the same thing?
Saihttam mentions that SMP and RAID are not well-supported in Linux. I'm wondering if this is an issue with OSS. Since most developers don't have access to high-end hardware, does this mean that features found in high-end hardware won't be as well-supported as those found in most consumer-based hardware?
Why would a company damage itself in such a way though? They appear to be using standard benchmarking tools, so is it possible that their findings are valid? I'm a big fan of Linux, but it doesn't mean that it is perfect. How many standard benchmarking studies have been done comparing the two? Much of Linux's performance ratings appear to be annecdotal. Not wanting to invite flaming, but just because it is MS, doesn't mean that it is immediately inferior.
Surely the more important question is *why* they're doing reports like this, rather than arguing about how much of the report is BS. Of *course* they lied, it's *marketing*. If they weren't at least a little bit worried about Linux/Apache they wouldn't need to fabricate the 'evidence', now would they?
Linux is by far the best, Head over to adnbench http://www.anvdesign.net/adnbench/ and check out the bench marks they have, It's a real world test on what normal people could buy and benchmark, Truth is that when linux is on a pentium celeron 300a, with 128Mb ram, and then you put nt on a celeron 300a with 128Mb ram, Linux kicks butt. Not only that but linux has out done it in the desktop stuff, KDE, Afterstep, and many others, Much more support for devices, I havn't been able to keep a nt box up for more than 4 months, But i have no problem keeping a linux box up for 2 years. I don't get this. why do you need to reboot just to turn on networking? Linux don't need to. Is nt lame? Or am i missing something? What about netware, Hell it's better.. :)
Got shack?
ShackCentral Network
Worlds best gaming network!!!
Isn't this MindCraft that one, who made an unfair comparsion test between NT and Novell Netware some time ago, and was heavily critizised by Novell spokesmen ? I am not sure, but remember so...