Linux Kernel Performance How Will 2.6 Measure Up?

An anonymous reader writes "This story offers some interesting performance comparisons between the latest stable Linux kernels (2.4.x) and the latest development Linux kernels (2.5.x), comparing performance on both a single processor and dual processors. These numbers help validate that the upcoming 2.6 kernel will outperform the current 2.4 kernel, at least in some instances..."

Comments

It's faster.

[Anonymous Coward]

Yes, It's faster for most tasks, especially interactive and high load tasks.

But how is this news? Ever since the thread on Kernel Notes a month or so ago, most of us have known it this.

Re:It's faster.

[The Analog Kid]

Well its just to put the restless at ease I guess, and maybe for better marketing.

Re:It's faster.

[Lebannen]

Linux marketing through Slashdot? Hehehe... next we'll have microsoft and intel buying banner space... Oh...

Re:It's faster.

[feepcreature]

Most of us?

I'm not sure that most of us read the threads on Kernel Notes... Many, perhaps... Some, maybe...

You overestimate us, I fear.

P.

Re:It's faster.

[Anonymous Coward]

Maybe I misread the numbers, but in quite a few cases the new kernel seems to be doing much worse than the old.

I guess that is the news then: "new kernel is not actually improved, it runs slower instead"?

Re:It's faster.

[Anonymous Coward]

Just one little victory, a kernel breaking free...

Re:It's faster.

[mackstann]

hm, and it already performed worse than the 'dying' BSD's.

Re:It's faster.

[mr_z_beeblebrox]

But how is this news? Ever since the thread on Kernel Notes a month or so ago, most of us have known it this.

Please, don't post "Look how cool and leet I am" and then not even sign your posts. Who mod'd that up to 4

Re:It's faster.

[Anonymous Coward]

"'dying' BSD's?!

You mean the Mac OS X series?

Re:It's faster.

[abirdman]

"...in quite a few cases the new kernel seems to be doing much worse than the old."

Thanks for saying that, I thought I had misread the data. It looks like they're still talking about it, and working on the code tweaks which would optimize performance. The referenced post (and following discussion) seems less than earth-shattering.

I've got to hand it to them, though: they're working hard for a good cause. And the better they do their job, the better it will be for all of the Linux users out here.

Re:It's faster.

[killmenow]

I'm not sure that most of us read the threads on Kernel Notes... Many, perhaps... Some, maybe..

Hell, most of us don't even read the articles posted on slashdot...

Re:It's faster.

[The Analog Kid]

Just one little victory, a kernel breaking free...

Must we quote Rush every other turn? Well I guess its better than Micheal Jackson.

Re:It's faster.

[Anonymous Coward]

Yes, It's faster for most tasks, especially interactive and high load tasks.

But how is this news? Ever since the thread on Kernel Notes a month or so ago, most of us have known it this.

I could have told you that 6 months ago, without any "kernel notes" nonsense.

Let's see:

- The O(1) scheduler
- The SMP IRQ lock broken down
- The preempt patches
- Lots of rewrites in bottleneck subsystems ...

Re:It's faster.

[Anonymous Coward]

I guess I just don't feel the need for a 733+ /. nick. It's the message that counts - not the messenger.

Compile time speedups

[IamTheRealMike]

The thing I'm most looking forward to is the better scheduling under heavy disk load. This'll hopefully make Linux a lot more responsive when compiling software, at the moment my machine can get bogged down and jerky when doing this.

Of course, the real solution would be to not need to compile software (plug plug :)

Re:Compile time speedups

[Anonymous Coward]

Yes, I experience the same thing running SQL Server on our Debian boxes here at work. I have been running a special MSDN members only version of the 2.6 kernel, and I must say, our Exchange performance has never been better.

Hats off to Linus and the guys...I have a Gentoo system here at home, and using the 2.6 kernel it happily compiles hard core VB operating system extensions while browsing with IE, with no slowdown. I love Linux!

Re:Compile time speedups

[sTeF]

why don't you use hparm -X?? -d1 /dev/hd?
that'll enable ata??/dma features. and everything will be much faster...

Re:Compile time speedups

[Arethan]

Autopackage sounds a lot like my pet project Linstaller. I stopped development a while back to get my CCNE and haven't restarted it since. One problem I ran into was what libraries you could expect to be installed on any given platform. Sure, there's the LSB, but does the LSB specify a base set of packages that make up a desktop or a server?

My aim was a little different from yours though. I was going for complete binary packaging from beginning to end. No source building, as automated ./configure; make; make install;s tend to make distro specific code. Instead I left the cross distro compiling up to the packager. All I provided was an archive format and a self extracting gui or command line installer that totaled under 50k of overhead. I stopped around the middle of implementing the scripting language backend. I didn't like the way it was going, and as I said earlier, CCNE was calling to me.

Maybe I should start it back up. It's not like I have much else going on lately. hmm...

Re:Compile time speedups

[Anonymous Coward]

Amazing, I've been running FreeBSD since 2.8 and I've never had an unresponsive system even while doing a build world; I guess the 2.4 kernel is alot worse than imagined.
Hopefully it will be solved for you Linux guys in 2.6.

Re:Compile time speedups

[cras]

I've tried 2.5.48 and 2.5.49 and gave up both mostly because when compiling software Galeon got horribly slow especially with scrolling, MUCH worse than 2.4 kernels. Giving smaller priority to compiling job made it better but I'm too lazy to type the extra nice command before make..

Re:Compile time speedups

[Luke-Jr]

Don't know what you're doing, but I can compile software fine w/o any slowdown in the UI. Only slowdown is in highly CPU-intensive apps like games or emulators.

Re:Compile time speedups

[pkplex]

Yep.

I first tried FreeBSD about a month ago, and thats exactly what I noticed about FreeBSD. Smooooooooth.

For example in Linux ( 2.2.x and 2.4.18+ ) I found that when something demanding was going on ( like building mozilla, kernel, or such like ), all X11 became all choppy ( Mouse stuttered, typing lagged in bursts ).

Not so with FreeBSD. Many times ive had GnomeICQ hit a bug and use 100% cpu, but I was unaware of this until days later when looking at top.

A few days ago I installed FreeBSD onto my p100, with 64Mb of ram. Playing around, I ran many many dnetc's by having thousands of 'nohup ./dnetc &' lines in a file and executing it.

At a load of 350 the p100 box was still very happy to do what I told it, and with very suprising responsivness. However, once the load got up to 450, my ssh connection to the box was terminated, and I had to restart sshd locally. Which is fair enough, I guess.. one will run out of swap and ram sooner or a later.

I can recall doing this same dnetc thiwith slackware, running 2.4.something, and after a short while at load 50, I started getting seg faults every time I ran a command.

Untill Linux shurgs off huge loads effortlessly and in a stable manner like FreeBSD does, its not going to live in my boxen :)

Tux needs to get fit and learn how to balance on one leg properly :)

Re:Compile time speedups

[IamTheRealMike]

One problem I ran into was what libraries you could expect to be installed on any given platform. Sure, there's the LSB, but does the LSB specify a base set of packages that make up a desktop or a server?

Nope, you're right, but autopackage can figure out what libraries are present and retrieve (assuming they've been packaged) the libraries from a DNS style distributed network, apt style.

My aim was a little different from yours though. I was going for complete binary packaging from beginning to end. No source building, as automated ./configure; make; make install;s tend to make distro specific code.

Hmmm, how did you get the impression that autopackage is source based? A .package is a binary package from end to end, the user doesn't need to compile anything.

All I provided was an archive format and a self extracting gui or command line installer that totaled under 50k of overhead

We're using a similar idea except the scripting language and front end code is external and installed-on-demand when you run a .package file if it's not already present to minimize package file bloat.

Maybe I should start it back up. It's not like I have much else going on lately. hmm...

If you're interested in the problem, please take a close look at autopackage first, feel free to hop onto IRC (freenode#autopackage) and talk to us first. We're normally around in the evenings GMT (both the core developers are in europe). It'd be a shame to duplicate effort when our projects sound so similar.

Re:Compile time speedups

[Bios_Hakr]

I know Mandrake has done this for a while. I think RedHat does the same. I can't remember with Gentoo, but I did try some hdparm flags and didn't notice any real change.

Basicly, do 'hdparm /dev/hd[x]' and look at the output. It will tell you which modes are in use for the current drive. Then do 'hdparm /dev/hd[x] -t' and see how fast your drive is running. Look at different optimize flags and test after each to find the best settings.

You can even use it to test cdroms and RAID arrays. Just remember that when you optimize an array, you want to optimize each disk (/dev/hd[x], not /dev/md[x]) seperately, but test the array as a whole.

One other note, the '-t' flag, like most synthetic tests, may not show the best settings for the drive. A lot of times a timed kernel compile (or my new fav test, a mozilla 1.0 compile) will reveal benifits, or detraction, not shown in a synthetic benchmark.

Re:Compile time speedups

[Uggy]

Maybe you should try
nice -n 19 make && nice -n 19 make install

No responsive problems AND your computer uses all those spare cycles while you are doing other things.

I do this all the time and desktop responsiveness doesn't bog down at all. Kernel doesn't take much longer to compile either.

Re:Compile time speedups

[sfe_software]

As much as I love Linux, I have to admit, I ran FreeBSD on a desktop for a while and it impressed the hell out of me. X11/Gnome and Mozilla ran at least 3x faster on the same machine, and under the heaviest loads it was still always responsive. It was the only system I ever trusted to burn a CD while I continued to do other things (under Windows and Linux, I tend to leave the box alone during burning).

Of course back in the 2.2 days, when 2.4 was on its way, 2.4 was touted as being better able to recover from heavy loads. We ran several Linux web servers, and a bad CGI (or a good /.ing) was too easily able to cause a downward spiral, ending in an inevitable hard reset.

2.4 improved that significantly, and even my home boxes noticed the difference. Another poster mentioned a new scheduler finally going into 2.6, so perhaps this will improve even further. Whether it will be as solid as FreeBSD or not, any improvement will still be great...

Re:Compile time speedups

[Anderlan]

Of course, the real solution would be to not need to compile software (plug plug :)

Plug for what???

I'll take the oppurtunity and plug Debian.
you@host:~# apt-get install practically-linux-program-youve-ever-heard-of

Re:Compile time speedups

[brsmith4]

What you describe definitely stinks of a mis-configured ide drive. As stated above, use hdparm (linux) to set the dma mode on your drive. This will greatly improve performance.

Re:Compile time speedups

[alexborges]

Try hdparm -m 16 /dev/hda .... I havent had slowdown problems since i enabled multiple sector io on my ide drive. Ive a big box though....

Re:Compile time speedups

[IamTheRealMike]

See my sig (autopackage.org). Debian package distribution has a number of issues that I won't go into here (about to go home) but I'll happily explain in IRC sometime or on the mailing list. I hope to improve on debians apt by quite some way.

Re:Compile time speedups

[jovlinger]

yes!

It sucks, but that is your fault. no, hear me out, this isn't one of those "so fix it" rants:

I've been building my own kernel since 1.2.13 days. I've [until recently] never built a crap kernel (sometimes left things out I wanted in, like sound, but it always worked smoothly, even under load).

I recently tried to roll my own 2.4.{7,18} kernels under RH 7.2, and it did exactly what you describe. The slightest bit of IO concurrent with load would stutter up the entire system.

However, redhat's kernels (based on the same version) would NOT have this problem. Smooth as astroglide on a banana peel.

So the conclusion is that the kernel broke sometime during the 2.4 task-switcher/vm mapper debacle, but not in a "no longer works" sense, but rather in that deep wizardry is neeeded to build a "good" kernel. Obviously you and I forgot to check the "do not fuck up" box.(*)

I would totally go for *BSD, but a clean RH install works ok, and I judge the overhead of applying updates to keep the system secure less than a complete OS shift and learning to administer a new not-quite-the-same system.

(*)The first instance of the "do not fuck up" box was found on the LaserWriter driver for System 7.x: if you unchecked download fonts, then _nothing_ would come out, otherwise it worked like a charm. Likewise, acrobat reader has an "avoid Level-1 PS like the plague" option that allows you to print what appears on screen even when it contains greek letters.

The problem is that each configuration dialog has this box hidden as something else, but it is always there, somewhere.

Re:Compile time speedups

[FooBarWidget]

They are both already turned on but everything is STILL unresponsive under heavy disk load.

Re:Compile time speedups

[chabotc]

What often helps a lot, is adding a -u1 (unmask irq). From the man page:
"A setting of 1 permits the driver to unmask other interrupts during processing of a disk interrupt, which greatly improves Linux's responsiveness"

Re:Compile time speedups

[yugami]

i read your faq, the main advantages (localization and interaction) that you list DO exist in deb's.

Re:Compile time speedups

[IamTheRealMike]

Yes, I know they do. The main advantages are not localisation and interaction, they are distro neutrality and decentralisation.

Linux has a big problem with software packaging. Deb files are debian specific. The answer is not for everybody to use debian, it's to build packages that are not specific to a particular distribution. If you never ever want software that's more up to date than the stuff in Debian, if you think dpkg is the essence of perfection itself and cannot be improved in any way, then my project is not for you. Sorry :(

For people who don't use Debian though (ie the majority) I think this system has major advantages.

Re:Compile time speedups

[Anonymous Coward]

Not everyone has ATA/DMA features, I for example have a quad PPro server whose onboard ide chipset will only run in 16 bit mode. Suffice it to say if it goes above the 1 meg/sec transfer rate the chipset can handle, the whole system is oWn3d until it finishes the disk access.

Re:Compile time speedups

[Ed Avis]

A well-written makefile should have the 'install' target depend on building the software to start with, so just 'nice -n 19 make install' should do it.

Re:Compile time speedups

[GrenDel Fuego]

For $50 or so, you can buy a UDMA IDE controller that will speed things up for you quite a bit. If this machine is running as a server, you might want to spend the bit of cash to get a big performance gain.

Of course, you didn't mention what type of server it is. If it's not a file or mail server, you might not care so much about disk I/O.

Re:Compile time speedups

[Anonymous Coward]

jesus christ now i know why my comments aren't modded up,they simply aren't understood..what kind of noob suggests that you use hdparm as if everyone has ide drives, or that everyone that has ide drives hasn't used hdparm for 'n' years.
Also this is really bad advice if your
mobo/chipset doesn't support dma, or if you have an old hard drive where dma is flaky. You can do serious damage/fs corruption if you enable dma indiscriminately without a clue.

Re:Compile time speedups

[Door-opening Fascist]

$50? For $20, I got a Promise ATA/100 card, which has functioned without a hitch for six months for me so far.

Re:Compile time speedups

[GrenDel Fuego]

Just a top of the head rough guess. I actually just ordered one for $30 myself in order to put a 40 gig drive in an old machine that only supported up to 32.

Quick question

[Anonymous Coward]

What is the weakest specced machine that anyone here is getting productive/useful work with Linux done on? Do people use Linux on 468s at 12mhz? P75s? Just curious.

Re:Quick question

[ChileVerde]

I'm using P133 w/ RH Linux 7.3. It works fine (very quick) unless compiling, and I even use X on it.

Re:Quick question

[Anonymous Coward]

There were 12 MHz 486's?

Re:Quick question

[Anonymous Coward]

What is the weakest specced machine that anyone here is getting productive/useful work with Linux done on? Do people use Linux on 468s at 12mhz? P75s? Just curious.

I am currently using a dual Athlon MP 2400 system with 4 GB RAM and a 10-drive RAID setup. I find performance acceptable, although I am looking to upgrade from my circa 1983 Hercules MGA card sometime soon...while the onscreen text is clean and crisp, I have found the preview of Doom III to be virtually unplayable.

Re:Quick question

[Anonymous Coward]

i am runing redhat 7.0 with a vanilla kernel 2.4.19 on a hp vectra 486 at 66hmz.

it is not the ultra fastest server in the world, but i use it as a router for my adsl/cable connection. and i am very happy with it.

Re:Quick question

[quigonn]

A friend of mine once had SuSE 5.1 on a IBM PS/2 laptop with 6 MB RAM and a 80 MB harddisk (or so). That was in 1998, I think. He used it for C development, writing TeX (only writing! not compiling, elvis' preview function was enough for him) and playing text-based games. He even had networking, via SLIP. If the power supply unit hadn't got a loose contact, he maybe would still use it.

Re:Quick question

[freedom_leffo]

Well, I'm still using my overclocked 486SX/25 (now at 33 MHz) with 16 MB RAM everyday. It's running a distribution built from scratch (http://www.linuxfromscratch.org) and it has got all the speed I need.

Re:Quick question

[Anonymous Coward]

My mail and news server is a 486... It only provides service for about fifteen people, so it's enough... Of course, it still uses the 2.2 kernel series being a Debian box that I don't touch. Doubt I'll ever bother moving it to 2.4, let alone 2.6

Re:Quick question

[neonstz]

386 SX-16 with 4MB ram. Was used as a tiny samba-server.

Re:Quick question

[BadDoggie]

486 or P75? Sure. Mail server. Firewall. File server. News server. Burner. CD library.

Hell, a P75 works fine as a Windows NT4 PDC for a small network and can also handle low-to-medium file serving for around 20 users at the same time.

Then there's the idea of using Linux network client stations, as in "How to create a Linux-based network of computers for peanuts", to which this site linked more than a year ago. This system can even make use of 386s -- I've already tried it. True, performance is a bit slack, but just how much power do you really need to write documents? A network-based 386 (or one running Slack 2.x) with Abiword or maybe pico/vi/emacs (some people do actually like those) works just fine.

woof.

Re:Quick question

[edgrale]

My experience with older hardware; if it's a Pentium you can use it as a firewall. I have a Pentium 133MHz / 64Mb ram / 4Gb hard drive. There is no fan, not even in the power supply. It's _quiet_ and doesn't produce much heat thanks to a LARGE heat sink.

I'm sure a 486 will do just as well. Btw. it's running IPCop (kernel 2.2.22).

Re:Quick question

[Dark Lord Seth]

I currently run Linux on some oddball Olivetti server with dual P100s. Works perfectly (so far, haven't tested the scsi tape drive yet.), even though I had to compile the kernel on my main machine. Would have taken 2 to 3 hours on one proc (SMP wasn't compiled in yet) which really isn't the most viable solution. Maybe I'll dig up a few 486, restore them, install some older stuff on it like the 2.0 or 2.2 kernel and try to sell them as cheap fileservers/webservers/routers?

Re:Quick question

[mvdw]

I have on my network at home the following ancient hardware:

• 486-DX 33 w/32MB RAM, running as dialup firewall for 2-3 users;
• 386-DX40 (with 387 co-pro), 20MB RAM, running slackware 8.0 (kernel 2.2.19 I think, although to be honest I don't turn it on all that often). I use this machine basically as a scsi box to format external scsi disks for my HP PA/RISC adventures.

I also have some other odd hardware on my network, that is so ancient that linux won't even run on it (2 DEC turbochannel machines, one MIPS-based and one Alpha, both running netBSD)...

I acquired all this hardware when it was being thrown out by various people - bang per buck is essentially infinite

Re:Quick question

[jabapi]

I used Debian 3.0 with Amiga 1200 / 68030@50 MHz / 32 MB RAM / 10 GB HD and it was running fine, except when installing or compiling new software. Recently I upgraded that to 68060/50 MHz and now even installing/compiling is fast enough.

And yes, that is just a test box for me. Naturally I have some speedier Linux boxes available but the good old Amiga still has its place. :-)

Re:Quick question

[teknikl]

I'm using a p100 with 32MB ram and a 3GB ide as my firewall host. Works great - compiles fast. Of course it's running OpenBSD which is somewhat less resource intensive than linux. Runs the NAT/PF - lets me ssh in. Not a problem. Previous firewall box was a 486/66 with 16mb - pulled off the PS fan and the case fan. Ran with no hardrive, booting from a floppy. Totally silent. Worked great.

Re:Quick question

[KjetilK]

I remember using a 386 20MHz running X and Netscape and lots of stuff back in 1995, when I worked for the student union. It wasn't fast but we could work on it. Eventually, we whined about it and got a new box.

Re:Quick question

[Zoolander]

Are you serious? Read the post again, with your 'funny spectacles' on...

Re:Quick question

[datadictator]

I was working at an ISP a while back.

We had a mail/rad/DNS sever running of a 486 DX 66 with 20 mb's of ram and a 200mb hard drive.

Distro was Debian 2.2, cut down to bae bones.

The catch - at any given moment more than 3000 people was reading mail alone - and the system remained fast.

No X btw. VNC,SSH and Webmin (for the junior admins) was all we used.

We upgraded it just before I left when the loads started going to 10 thousand concurrent users.

Re:Quick question

[Anonymous Coward]

No, but the very first 386s were 14.7MHz. The initial target was 16MHz. They followed soon afterwards. I remember when the 386/25 came out :-) Ah, those were the days :-)

Re:Quick question

[Anonymous Coward]

At my last place we had one of our leased-line company firewalls runing on a 486/25 with 4 Mb memory and 200Mb, Slackware 2.1. We had ten times more trouble keeping our state of the art CISCO routers going, than that little miracle. In 12 months it only crashed once!

Re:Quick question

[the_real_tigga]

Well, we've got kind of a higher-end low-end system here, a dual P90 with 64MB EDO. works great as mail and firewall/masq server, and with about 10 users it can also handle some file serving/sharing, plus rsync backups, cvs and CD burning (4x, SCSI).
Plus, with this crate I learned wha the "EISA Bus Configuration" in the linux kernel is for...

I have also had a 486DX/2 100 for the same purpose, ran great too, but had to decommission it because it wouldn't take larger HDs.

(On a side note, I was able to recycle the 30-pin SIMMs from the 486 by putting them into the cache expansion slots on the P90s EISA SCSI controller card. I miss the old times.)

Re:Quick question

[Anonymous Coward]

Sun SparcStation 5 with 32Mb RAM, 6Gb SCSI disk, 4mm DAT and Linux Debian. It's used for mail server (only text e-mails), file server (just archive of unused docs, low load) and backup. Works great, several months uptime and no problems at all. Best machine I've got.

Re:Quick question

[Anonymous Coward]

What is the point to VNC since you dont have X on the machine?

Re:Quick question

[Anonymous Coward]

I'm running Astaro, a Linux based firewall/router distro at home.

It runs on a P133 w/32mb of ram and a 5gb hdd.

It's not super fast, especially when I'm trying to use the web configuration client, but as far as being a firewall, and for cli, it's fast enough...

Re:Quick question

[Anonymous Coward]

> What is the weakest specced machine that anyone here is getting productive/useful work with Linux done on?
> Do people use Linux on 468s at 12mhz? P75s? Just curious.

Newer distros assume you have at least some hundred megabytes of disk space, a decent amount of memory and a relatively fast CPU.
To use old hardware efficiently you should look at projects like uClibc or Busybox that are intended for embedded/small devices.

Re:Quick question

[vofka]

I have a 486SX/25 with 12MB of RAM (2x 4MB SIMM's and 2x 2MB SIMMS), and 750MB HDD (1x 250MB and 1x 500MB)...

It does:

• Slackware 8.0
• 'net Dialup
• Firewall (IP Tables, 2.4.18)
• Mail Server (Sendmail, both in and out)
• Web Server (Apache)
• Internal Forwarding DNS Server
• Distributed.net Proxy
• Distributed.net Client
• Webcam Capture
• SSH Tunnel / VPN Server

And it still has enough spare capacity to do compiles (though the kernel takes about an hour and a half!)

Re:Quick question

[Alien Being]

"and doesn't produce much heat thanks to a LARGE heat sink"

And I don't get fat thanks to a LARGE stomach.

Re:Quick question

[Anonymous Coward]

I have a 486/25mhz laptop with 8mb of ram and a 50mb harddrive manufactured by GRiD computing. It is a ruggedised laptop so is worth keeping around (I've dropped it down a flight of stairs without a scratch). It is currently running Slackware 3.6 with a 2.4.19 kernel(Much faster than 2.0!). It has a PCMCIA slot, so I use it daily as a wireless color terminal around the house. It runs X(8bpp) nicely, SSH, EPIC, mICQ, and various other small applications. I also use it to compile libc5 binary packages on occasion.

Re:Quick question

[marco_craveiro]

my main pc at home is a toshiba satellite p1 100 mhz with 16mb RAM and a hard disk of around 800 MB. i used to run X on it but it took a while to start so now i use command line only. i run debian with a 2.2 kernel and i have used it to code and write my thesis (c++ and TeX) without any significant problems. i browse the web with lynx and read mail with emacs. i also run mono on it (learning .net).

Re:Quick question

[dennism]

I host my personal web server on a P120 running Linux 2.2 on some version of RH (doesn't really matter now with all of the other upgrades on the box). It only has 32megs of RAM, but it's happily serving web pages, a SSH server, a MySQL database, and all of the other services that I use from the inside of my network (telnet, ftp, atalk, etc).

It's been running for years now without any problems -- it could probally use a bit more memory, but then again, swap space hasn't even been touched at this point.

All in all, a very good box -- especially considering all of the power outages/brownouts that I've had the past few years. This box just keeps on going.

Re:Quick question

[Anonymous Coward]

Please, link me to a place describing such a setup!!! I have a spare 486-33 to play with.

Re:Quick question

[dasunt]

My weakest spec machine I regularly use is a P75 laptop. 16M memory, 750M disk.

OS is Debian Woody.

Desktop is XFree86/Ratpoison.

Primary programs are ssh, w3m, centericq, and dillo. (Although I'm moving more towards 'View' in w3m, as well as email and newsgroups in mutt).

Why? Because its nice to have a cheap sturdy laptop I can take anywhere without worrying too much about it.

I also use a P166 at home for a fileserver/mailserver/newserver/sambaserver/firewa ll/dialoutserver.

Re:Quick question

[Anonymous Coward]

Only a dual CPU? That's your problem. If you go quad, you devote the third processor to beefing up the Hercules card to acceptable levels. RTFM. Sheesh! N00b...

Re:Quick question

[Phreakiture]

I have, currently in service, a K5 running 133MHz with 64MB RAM and 4GB disk. It is being used primarily as a near-real-time dynamic range compressor and audio recorder. I say near-real-time because there is approximately .5 second lag between the time audio enters and exits the box. It is able to keep up without damaging the audio.

Also currently in my service is an AMD 5x86 (a glorified 486DX4) running 133MHz with 32MB RAM and 4GB SCSI disk. It is the border box for my home network, and works as a web server (Apache), web proxy (Squid), ssh server (OpenSSH; for using the web proxy and accessing internal machines remotely), and does NAT and some firewalling. I am also going to add a POP proxy soon for spam filtering purposes (POPfile).

A friend of mine has two web servers, one is a 486DX running 33MHz with 16MB of RAM and 2GB of SCSI disk. He also has a Sun Sparc IPX (40MHz, I think?) with 32MB of RAM and 4GB of SCSI disk.

Re:Quick question

[idontgno]

My current household IMAP/Appleshare/logging/SMB/dialup firewall machine is a P133, 96M. That's after 3 motherboard swap-ups, as other machines in the household got their MB upgrades. The box started out as a 486 DX/4 100 with 32M and VL-bus slots. (Old school!) And all of it in an old AT-style desktop case on edge and wedged tower-style under the corner connector in my computer desk. (Fits good into the narrow space between desk and printer stand; a real tower is too wide.)

Built out with RH 7.2 but no X. (It's a server; I don't even turn on the monitor most of the time.)

Works great, but the last 2.4.x recompile took over an hour. Anymore, I don't bother with source compiles and just let Red Hat Network update my stuff. (Lazy administrator!)

Re:Quick question

[cronus42]

I have 3 systems running Debian Woody on custom compiled kernels (2.4.20-Final).

#1. AMD Thunderbird 850Mhz, FIC AZ11, NVidia Geforce2 MX, 512MB PC133, 11.5Gb IDE HD, 32X CDROM, 4X4X24 CDRW, 56K Dial up w/ 10baseT local ethernet.

#2. IBM Thinkpad 360CSE, 486DX2 50Mhz, 20MB DRAM, 340MB IDE HD, DockII w/ 4X SCSI CDROM, 10baseT Ethernet.

#3. IBM Thinkpad 360CS, 486SL 33Mhz, 20MB DRAM, 340MB IDE HD, Same networked Dock II.

I have no problem with any of these systems. I use the AMD system with diald and ipmasq/dnsmasq as a gateway for the networked dock. (YES! I really share a 56K connection. The laptops don't have X, so text only web doesn't need much bandwidth.)

Re:Quick question

[Anonymous Coward]

I used a 486DX/2-66MHz as a cablemodem router/firewall for over a year with no speed problems. I have since upgraded to a P200MMX. I use another P200MMX as my main GNU/Linux X-windows programming PC.

G.

Re:Quick question

[Anonymous Coward]

3000 users on a 200MB disk? Did each user have a 50KB mail quota?

Re:Quick question

[Anonymous Coward]

I've got it running on a rather slow 486 with 16MB.
It has over a year of up-time. It is used as a
controller for an astromonical camera in a "lighs
out" enviroment (no pun, I mean the CRT has not been
powered up in over a year.) It runs a version of
Linux modified for real-time.

I have an even lower end machine: A Sum IPC. with 12MB RAM. This is a very slow 32-bit SPARC CPU
I think the IPC runs at 20Mhz and is about equal
to a 20Mhz Pentium (if one existed.) It worked
OK as a low volume server.

I've also tried Linux on an old 68K mac with
something like 8MB RAM. I think it was an
Ci or Cx. A foot square box about 4" tall.

Re:Quick question

[Anonymous Coward]

Have you tried using aalib to render it on the console? :)

Re:Quick question

[mark_space2001]

Well, I'm using a P166 w/ 32 megs of RAM as a firewall and NAT. And I'm only running the shell, no GUI. With X, that system would barely run, but with just the shell, it's really fast.

I found RAM to be the limiting factor, rather than CPU speed. Thirty two megs is the minimum Red Hat recomends for it's 7.X system (which is what I have). The kernel plus modules and buffers seem to take up most of the available memory (according to /proc/meminfo). So I wouldn't want to try it with much less. And for networking, you're much better off if you don't have to swap stuff out to disk just to route a packet.

I think you could get it to work with 16 megs but less I'd be worried that performance would truly suck. Anyone know different?

Re:Quick question

[RealAlaskan]

What is the weakest specced machine that anyone here is getting productive/useful work with Linux done on?

I have a 486 (75 MHz, I think) laptop, with 12 MB RAM. It runs Debian Woody with X11 and the Blackbox window manager. I was using it yesterday to read slashdot and debianplanet. I was using the Dillo browser. Performance was slow, but tolerable. If I want to use emacs, I kill X, or else it starts swapping.

The real holdup for speed is the lack of RAM, rather than the anemic processor, but as long as I'm careful about how many processes I have going, it's fast enough. The holdup for usability is the small, 640x480 screen, and that's more a matter of it being an old laptop than being old.

It's interesting to notice that this little box could run Win3.1, but isn't a speed demon with that either. Win3.1, of course, is old and single-tasking and insecure and nothing modern runs on it. Putting a modern, proprietary OS on this sort of old hardware is probably out of the question.

Based on this, I think that any Pentium with a lot of RAM (more than 64 MB) would be just peachy for email/websurfing. Any Pentium with >32MB should be useable, if set up sensibly.

Re:Quick question

[Anonymous Coward]

you should upgrade to Slackware 8.1.

Mike Cho

Re:Quick question

[Anonymous Coward]

With X, that system would barely run, but with just the shell, it's really fast.

You're using the wrong X environment. I run X with fvwm2 and even mozilla(!) on machines punier than that. Just don't waste your time with GNOME or KDE and whatever crap Red Hat decides to ship with your distro.

Re:Quick question

[AJWM]

I have a 486DX-66 (33.18 bogomips, 24MB RAM) that I'm still using as a mail server. I used to also run 3 web sites and DNS on it, but since moved those to a different machine. (More a disk space issue than anything else.)

I'm also running a SparcStation IPC (24.88 bogomips, 32MB RAM) as an internal web server and CD image server. (Running SuSE SPARC 7.3).

Up until a few months ago my development machine at work was a P-166. (My main home machine is a dual CPU P3-550 with a half gig of RAM, and yes, I noticed the difference!)

Re:Quick question

[Anonymous Coward]

I use a p100 with an old TX board , an old 4.3 gb wd ide drive, two nics(one ISA, one PCI) and 32 megs edo ram that I rebuilt from pieces years ago.

It has run for two years now serving as my dns server(two small zones internal(one kerberos subnet, one "open" subnet) plus all the usual inet lookups), print service (lpd, internal),an internal ftp server(vsftpd), apache 1.3.26 for cgi development and web doc reference,assorted small homebrewed services,(an ascii pokergame server, a snortlog server, a couple honeypots) and router.


You'd have to see it to believe how nicely it works.

Re:Quick question

[inode_buddha]

I have a 486SX running X with FVWM on 20 megs of ram in one corner, a P75 with 32 megs (no video, keyboard or mouse), and a Compaq Deskpro 4000 with 96 megs and a Trident video card running RH6.2 with *lots* of updates. My backup server has dual PPros and a lot of disk since a decent tape drive is just too expensive. I standardized on kernel 2.4.19 with the (0)1 scheduler patch for all my boxes, and built it from tarballs for each specific CPU type on my main workstation. (Main workstation has dual P3 Coppermines at 1 Ghz and 1048 megs of ram.) All of these machines can do the usual linux office apps (NOT StarOffice -- it uses too many resources). You won't do too much multimedia on them tho, and a kernel compile takes 8 hours on the 486. That's why I have the new workstation mentioned above.

Re:Quick question

[Anonymous Coward]

Fuck knows. I can't remember that far back. There were some sort of 12mhz machines then. People liked to buy games which ran too fast on them - because they could show off how fast their pc was- rather than at the right speed. I heard this from a games developer!

Re:Quick question

[Bert64]

What`s the point of using VNC atall if you have X ?
X supports remote use natively, no need for a hack like vnc, and its sure a hell of a lot faster.
Whats more, X natively lets you export single apps at a time, without the overhead of a complete desktop and windowmanager, and the apps from one server integrate perfectly with the ones your running locally.

Re:Quick question

[Bert64]

X used to run well on an identical machine not so long ago, pity that modern toolchains (libc etc) and X is so much bigger than it used to be..

Re:Quick question

[jbridge21]

censored.firehead.org is a dual P100 with 32MB RAM and a decent SCSI subsystem (one disk). It gets many many hits a day.

Re:Quick question

[Eivind]

Depends on what you use Linux for. I've set up a packet-filtering router for a 3Mbps wireless link using a 386 computer and 8MB or RAM. The entire thing boots from a floppy, and runs everything from a ramdisk.

Works fine. The hardware is more than adequate for shuffling a few Mbps from one nic to another while inspecting some ip-headers.

It may be faster...

[cperciva]

but will it corrupt my filesystem?

Performance is important, certainly, but I think some people (*cough* overclockers *cough*) assign it a bit too much importance.

Re:It may be faster...

[Anonymous Coward]

Performance is extremely important. Overclockers are stupid however. Actually I'm slightly Oced here -- it's a long story.

SMP

[e8johan]

It looks like the new kernel better utilizes multiple CPUs. This is a great thing. Linux needs better support for SMP systems if it is going to play with the big kids in the high-end server market. (I know, Linux is partially there).

Will 2.6 make servers...

[Negatyfus]

more resistant to the /. effect?

Re:Will 2.6 make servers...

[Clay Pigeon -TPF-VS-]

How is that off topic? Hes asking about performance under Extreme loads. That is if hes being serious. If hes not then we shall crucify him! I wonder what this will do to my karma...

Re:Will 2.6 make servers...

[Anonymous Coward]

Well.. they are including among other things epoll() which should scale well for thousands of clients. It serves the same purpose as select() and poll() but works a bit different.

Of course that doesn't going to help www-sites that want to run a lot of dynamic content for every request.

Sorry to Intrude here, but...

[krystal_blade]

sic"Expect the 2.6 to outperform the 2.4, at least in some instances."

Really... I mean, that is the whole point to actually making an IMPROVED product, right... So it's an IMPROVEMENT over what was offered before?

I know, I know... Don't beat the messenger...
But, I swear, if a company uses that line to sell a linux distro, I'm going to beat their marketers.

Good selling point. Well, yeah, it does ok, I guess... In a few places...

krystal_blade

Re:Sorry to Intrude here, but...

[feepcreature]

Really... I mean, that is the whole point to actually making an IMPROVED product, right... So it's an IMPROVEMENT over what was offered before?

Hmmm. Microsoft (and many other software producers) tend to improve by adding more features and reducing performance. So this is actually quite a revolutionary concept.

Of course, marketers will need to come up with something that sounds more impressive than "outperform the 2.4, at least in some instances".

P.

Re:Sorry to Intrude here, but...

[Xpilot]

The whole point of Linux development is to , explore strange new algorithms, seek out new drivers and new filesytems, to boldly code where on one has coded before :)

As with all experimental endeavours, you do sometimes get better results, sometimes worse, but from those mistakes lessons are learned and better methods are devised.

It's not about "marketing". It never was.

Re:Sorry to Intrude here, but...

[Simon Kongshoj]

cheer!

Why does it seem that so many people in the current Linux community *think* that it's about marketing and money, though? *sigh*

Re:Sorry to Intrude here, but...

[Anonymous Coward]

Oddly enough, M$ has made absolutely no effort to improve performance with each successive release... quite the opposite, in fact!

Re:Sorry to Intrude here, but...

[xanadu-xtroot.com]

people in the current Linux community *think* that it's about marketing

Because a lot of the new-comers have been beat-up with Billy G's ugly-stick for so many years. That's all they know. Marketing ploys and false hype.

monkeyboy anyone?

Re:Sorry to Intrude here, but...

[Simon Kongshoj]

I'm afraid part of it might be because of the Linux community itself, though.

After RedHat and VA made it big, way back when, a certain amount of "make money fast!" thinking crept into the Linux community. It started seeping into our news and other "internal" communications too (I mean /., Newsforge, Linux e-zines and that sort of stuff, not the lkml) -- people started focusing more on how one could use Linux and open-source to make profit rather on technological issues. The line between hacker and marketer seemed to be breaking down. A O(1) scheduler is all fine and dandy, but how can my favourite business use it to make more money?

Perhaps I'm seeing something that isn't there -- I hope I am -- but seriously folks, am I the only one who have noticed such a switch of focus? I read the articles at NewsForge about the last LinuxWorld Expo, Roblimo seemed to agree that the old bunch of long-haired hackers in sandals had largely been replaced with business reps -- and to add insult to injury, Microsoft was present.

My fear is that Linux will end up becoming as sterile and dead as other "rebellious" technological (or otherwise) ventures tend to become when they're subjected to corporate clutches. Greed kills.

Re:Sorry to Intrude here, but...

[Sj0]

With Linux 2.4, the internet will be faster!!!

j/k

Re:Sorry to Intrude here, but...

[inode_buddha]

No, you're not the only one who has noticed the switch in focus. I know a few large engineering companies that could use the (0)1 scheduler and save huge amounts of time (which equals money), but their management has standardized on MS products at most levels. Also I'm glad I have a gray beard and long hair, sometimes. It's good to be able to reminisce and hopefully learn from the past while trying to create the future. Though I prefer combat boots, not sandals.

BSD?

[pkplex]

I wonder how well it compares to the BSD kernels, in both performance and stability?

Re:BSD?

[Xpilot]

I wonder how well it compares to the BSD kernels, in both performance and stability?

I was wondering that myself. The *BSD's are more or less "distros" complete with userland tools whereas Linux is just a kernel. The complete OS+tools should be called GNU/Linux (as RMS insists).

Whenever I see posts that say "*BSD is better than Linux", most of the time they are referring to some userland aspect of *BSD compared to some GNU/Linux distro, and not "Linux" the kernel itself. Which isn't really fair IMHO, since most of the time it's not "Linux"'s fault for whatever is being bashed at the time.

Re:BSD?

[Cid Highwind]

Which isn't really fair IMHO, since most of the time it's not "Linux"'s fault for whatever is being bashed at the time.

Score: -1, Bad Pun

Re:BSD?

[Anonymous Coward]

I dunno, but at least Linux isn't dying.

Re:BSD?

[Daniel Phillips]

Whenever I see posts that say "*BSD is better than Linux", most of the time they are referring to some userland aspect of *BSD compared to some GNU/Linux distro, and not "Linux" the kernel itself. Which isn't really fair IMHO, since most of the time it's not "Linux"'s fault for whatever is being bashed at the time.

Not exactly right. There's no question it was very much Linux's fault for having a less than totally robust virtual memory manager for a number of years. In the push to add features such as memory above 4 gigabytes, stability in corner case and swap performance kind of got left behind. This has been corrected in Linux 2.5 with the new reverse-mapped VM, which sacrifices a little raw speed in such things as process forking (look closely at the benchmarks and notice 2.5 is slightly slower in Con's "process load" benchmarks) and mallocing, in return for far better and more predictable swapping performance. Plus, the new VM provides a better base for new developments you'll see in the next series, such as active memory defragmentation. Over time, we're likely to win back the slight performance losses in (certain areas of) the 2.5 vm, and then some. In the meantime, there's no question that 2.5 is the smoothest running Linux kernel ever.

BSD continues to edge out Linux in some areas, notably NFS server performance. It used to be, BSD had a lot more advantages over Linux than it does now (the BSD developers are darn good). But in the end, Linux offers a much broader range of hardware support and has way more programmers working on it, so slowly but surely is catching up and surpassing in the few areas where BSD still has the edge. If I had to speculate about why Linux gets the massive herds of programmers, I'd say it's because of the license - many volunteer programmers prefer the GPL because of the legal guarantee that their work will remain open and not end up fading away because it had to compete against some heavily-funded proprietary product based on their own code. However, it's clear there are enough top-flight programmers to whom such considerations are unimportant to keep the BSDs not only alive, but vibrant.

See here for a look at some of the nice features BSD, and some ideas for the future. In case anybody thinks the much-talked-about rivalry between Linux and BSD is some kind of war, it isn't. BSD and Linux people often work together, there is a lot of cross-pollination, and the prevailing attitude is one of mutual respect. At the end of the day, it's worth noting that, technically speaking, the closest rival to Linux in the operating system space is another open source project.

Re:BSD?

[Foresto]

"The complete OS+tools should be called GNU/Linux (as RMS insists)."

Hogwash. Adding "GNU/" when communicating on the subject of Linux does absolutely nothing to improve communication. It's well known that GNU tools are used with every common distribution of Linux. Prepending additional letters and symbols to the word "linux" needlessly complicates communication. I'll consider doing it when non-GNU tools become common enough with Linux distros to cause an ambiguity.

Re:BSD?

[Arandir]

Adding "GNU/" when communicating on the subject of Linux does absolutely nothing to improve communication.

The correct name is Grub/GNU/Linux.

Re:BSD?

[Bert64]

So what next?
Intel/Maxtor/AmiBIOS/Nvidia/Via/Logitech/Gr ub/GNU/ Linux ?
Afterall, the hardware is important too! And dont forget the firmware.
Which makes me wonder, why dont we have OpenBOOT/MACOSX, or OpenBOOT/Solaris, or maybe SRM/OpenVMS ? why not? because its stupid.. having a short name is much easier for people to refer to.

Aside from that, many people dont use grub to boot linux, and many other people use grub to boot other os`s such as windows.

Re:BSD?

[Arandir]

If you would have followed the link, you would have caught on to the satire.

P-75s

[lprechan]

P-75 w/32Mb RAM firewall
P-75 w/32Mb RAM printserver for 3 node network. Sure, small time lag when sending graphics to print server but overall pretty nice job.

not just desktops

[g4dget]

People don't just use Linux on desktops, they also use it on handhelds, wearables, and embedded systems. So, a 486 running at 12MHz isn't out of the question at all.

SMP is overrated

[g4dget]

Yes, there is some market segment that really swears by SMP. But beyond dual processor machines, the hardware cost and engineering complexity grow disproportionately with the number of processors. Most of the SMP market is driven by companies facing excessive software license costs for multiple machines, or by companies that can't figure out how to get their current architecture ported to a cheaper distributed system.

When it comes down to it, you only get cost-effective scalability by using distributed systems or clustering. In fact, for really large systems, it's the only possible way at all.

Something like OpenMosix should really be a standard part of the Linux kernel already, as should other support for simplifying clustering, distributed computing, communications, and distributed shared memory. Distributed systems and clustering is the future, not SMP.

Re:SMP is overrated

[e8johan]

SMP still has advantages if a number of processes needs to share lots of memory. For other, more suitable, problems, mosix rocks!

Re:SMP is overrated

[g4dget]

With Gigabit networks on the one hand, the cost of many-processor SMP systems, and cache effects, it isn't clear that there are any problems for which SMP is the most cost-effective choice to achieve a given level of performance.

Mosix, of course, is no substitute for the kinds of problems in which many processes share a lot of state, but other approaches are (including distributed shared memory and various other communications libraries).

Re:SMP is overrated

[oliverthered]

Your missing out:
Latency, Mosix is just too un-responsive compaired to an SMP option.
Time, The more people that buy SMP boxes the better they will get, the MHz wars and Windows killed of home SMP if Intel had invested in SMP design instead of MHz then there'd be cheeper cooler SMP machines out there.

Re:SMP is overrated

[rikkus-x]

SMP is not overrated for CPU intensive tasks.

For example, I can do distributed compilation and get about 140% speed with 2 identical machines, but with SMP I can get more like 180%. It's cheaper to buy an SMP motherboard, 2 CPUs and some slightly more expensive RAM than a whole other machine.

Dual Athlons are great, price/performance, for compiling large C++ projects (where g++ needs lots of CPU for each file.)

Rik

Re:SMP is overrated

[Daniel Phillips]

When it comes down to it, you only get cost-effective scalability by using distributed systems or clustering. In fact, for really large systems, it's the only possible way at all.

Three years ago you would have been right, but today the cheapest way to (nearly) double your computing power is to put in a dual processor board. I.e., the day of the home dual-processor has arrived. For example, you can now get a dual processor Athlon board for $200, and in spite of what the docs say, you can put $50 processors in it instead of the $500 big brothers AMD recommends.

It's only a matter of time before you start seeing 3D games that can take advantage of dual processor configurations. In fact, they already can in the sense that if a single-threaded game can load up one processor 100% and your box still remains entirely responsive for other applications. That is, you can play Return to Castle Wolfenstein at the same time you run a compile.

Re:SMP is overrated

[dmelomed]

Not everybody runs big SMP servers though.

NUMA is more scalable and less bottlenecked than SMP for big servers.

Re:SMP is overrated

[g4dget]

Latency, Mosix is just too un-responsive compaired to an SMP option.

Latency depends on the problem and the design of the distributed system. For problems where Mosix is an alternative to SMP, latency doesn't enter into the picture at all because Mosix processes are usually only loosely coupled and because network and disk I/O migrates with the processes.

For problems where IPC latency is a performance concern, it can almost always be dealt with even in a distributed system through better design. The money you save on overpriced SMP designs more than lets you make up for any remaining performance losses from network latency (and I'm not convinced that with modern networking technologies, there are significant losses anyway).

The more people that buy SMP boxes the better they will get,

No, they won't. There is no magic tooth fairy of SMP, and you can't scale SMP indefinitely. If you put, say, 64 processors onto "the same memory", they aren't really on the same memory anymore--you are just paying for a very expensive box with a bunch of hamstrung processors inside that are essentially doing distributed shared memory over a special-purpose network.

Re:SMP is overrated

[Anonymous Coward]

No shit. I tried to compile qt from source with a k6 550, k2.4.12, older glibc, I could've used 2 or 3 additional processors anytime.

Re:SMP is overrated

[oliverthered]

"you can't scale SMP indefinitely."

Well, your kind of wrong.
X86 is crap for SMP, mainly because of the way memory and cach are handled. I wouldn't put more than 4 x86's in an SMP configuration.

Other architectures support point to point busses and better cache handeling more memory chanels/controlers etc... you could probably scale to 100 or so processors with that type of architecture.

Now, if some of the patents cray has were available then you can scale to thousands of processors, where memory controlers record pages that have been writen to and only sync on demand (when there's a read request page that's been writen to by another processor)

Basicly x86 sucks for SMP (which is probably why it's such a cheep architecture).

Re:SMP is overrated

[g4dget]

Other architectures support point to point busses and better cache handeling more memory chanels/controlers etc... you could probably scale to 100 or so processors with that type of architecture.

That's an illusion. If there is any significant per-processor caching going on, you basically have a distributed systems with a fast, non-standard network in between, and a costly, complex page fault handler hardwired in hardware.

You can achieve the same thing more cheaply with distributed shared memory and a standard fast network. That is, instead of building lots of expensive, inflexible, special-purpose hardware, you treat the main memory of each machine as the "per-processor cache" and you do the synchronization in software over the network using page fault handlers. It simply makes no economic sense to put something that complicated and costly in hardware, in particular if the market for it is so small.

Note that architectures like MIPS already do page fault handling in software, so those kinds of software-based approaches are competitive with hardware implementations.

Re:SMP is overrated

[autechre]

Or, when your one user that insists on running rtin on the server quits, and rtin spirals upward to 99% CPU usage, the machine is still usable by everyone else because rtin is only pegging one CPU. Replace rtin with any other program that does that (though the kernel is getting good at murdering rogue processes).

Re:SMP is overrated

[oliverthered]

'particular if the market for it is so small'

Which gets back to my first point, if more people used SMP then there would be better cheeper SMP system out there. since SMP isn't a consumer product the only kit out there is very expensive.

As for performance, compaire the ASCI programs with the Earth simulator.

People in glasshouses shouldn't throw stones

[marm]

Amazing, I've been running FreeBSD since 2.8 and I've never had an unresponsive system even while doing a build world; I guess the 2.4 kernel is alot worse than imagined.

This is, by and large, the fault of the scheduler, largely unchanged in 10 years and described by Linus, even whilst he wrote it, as a 'hack'. However, it worked, and Linus, being the extremely sensible and conservative maintainer that he is, kept it until recently - process schedulers are difficult things to get right, and their performance is crucial to the performance of the kernel as a whole. Not to mention that for the tasks that Linux has been used for historically, primarily low-volume server tasks on low-end hardware, it isn't really a bottleneck.

Still, the scheduler has been gutted and rewritten for 2.6 by Ingo Molnar - the now somewhat-famous O(1) scheduler, which performs much more fairly under load, and dispenses with almost all of the strange pauses and scheduling glitches under load. Current vendor kernels based on 2.4 (Red Hat's and SuSE's at least, I think) have had the O(1) scheduler backported to them as well. In fact, if you're running near enough any current 2.4 kernel other than mainline, you get the O(1) scheduler and your share of scheduling fairness.

The new scheduler is also a fundamental basis for Linux 2.6's new NPTL 1:1 threading, which has so far proved spectacularly (record-breakingly?) fast. Hmm, on second thoughts, perhaps I probably shouldn't mention threads and FreeBSD in the same post. I mean, isn't this the same FreeBSD that's still waiting for a single half-decent pthread implementation? Oh well, better hope 5.0 is out soon...

Re:People in glasshouses shouldn't throw stones

[Anonymous Coward]

It's nothing wrong with the POSIX threads implementation, its perfectly adequate, it's just that the FreeBSD scheduler don't do on the internals of the userprocesses.
Unlike other implementations FreeBSD run kernel and userspace threads. If you run userspace threads (POSIX threads) and expect scheduling you're out of luck, unless you write a userspace scheduler for the process in questing, which is quite trivial.
Otherwise you can just use the thread libraries in the ports tree which does that perfectly.

Re:People in glasshouses shouldn't throw stones

[marm]

It's nothing wrong with the POSIX threads implementation, its perfectly adequate, it's just that the FreeBSD scheduler don't do on the internals of the userprocesses.

Ok, let's get this straight. FreeBSD's default pthread implementation is an entirely userspace affair, everything occurs within a single process and the thread library does co-operative scheduling of threads within the process. Not quite sure what you're getting at when you talk about writing your own thread scheduler, the pthread specification requires that the thread library schedules and run your threads for you. A thread library that doesn't make sure threads get scheduled isn't really a thread library anyway, is it?

It works, just about, I'll give it that. However, I wouldn't call it perfectly adequate, it sucks donkey balls. It is slow, inefficient, and prone to threads blocking the whole process. It doesn't take advantage of SMP. You link to the re-entrant libc version, libc_r instead of the normal libc, and libc_r is incomplete, which means that quite a lot of pthread-using code that works fine on other platforms craps out on FreeBSD. Forget about realtime threads - you can attempt to create them, but without kernel assistance they can never be truly realtime, so it's not really a full pthread implementation anyway. Worst of all, it's buggy - it doesn't take a lot of load for very long for things to start behaving erratically.

Ok, that's just the default FreeBSD pthreads, and there are better alternatives - ironically probably the best is LinuxThreads, the standard Linux pthreads before NPTL, which attempts to do 1:1 kernel threading. It seems to work ok, but it's a whole lot slower than it is on Linux, because there's no screaming-fast clone() syscall like there is on Linux and precious little other kernel assistance. It's a little buggy too, not suprising given that it's not running on the platform which it was designed for, although it's bearable. Having to link to something in ports just to get halfway-usable threads is a really dumb situation for FreeBSD to be in, and a total PITA for developers who want their threaded software to run on FreeBSD too.

This ought to be all fixed in FreeBSD 5.0, it's getting a new pthread implementation using KSEs (Kernel Scheduled Entities), basically M:N threading like most modern OS'es use (and which Linux NPTL appears to have just obsoleted as a concept ;), but it isn't finished yet and FreeBSD 5.0 isn't out. Meanwhile, threading in FreeBSD 4.x is still dreadful.

Re:People in glasshouses shouldn't throw stones

[inode_buddha]

Funny thing that you should mention the (0)1 scheduler - I think it's one of the best things out there. Both my boxes are SMP, with RH8 and a homebrew 2.4.19 kernel and the scheduler patch. A typical workload includes a few concurrent compile jobs, reading /. with mozilla, and enjoying some mp3's - all at once. The older box performs without a hitch - it's dual PPros with 1Mb caches and 128 Mb ram. It keeps up with my Dad's new P4 on XP for all the usual office-type apps and stuff. I haven't been able to get my head around the NPTL yet, though I'd love to get that going. Any good HOWTO-type links for NPTL? Thanks.

Re:People in glasshouses shouldn't throw stones

[Nexx]

basically M:N threading like most modern OS'es use (and which Linux NPTL appears to have just obsoleted as a concept

Please explain why M:N threading is obsolete. For example, I don't see why the kernel should devote kernel resources to a thread that is blocking on IO.

So you overclock to go faster...

[Rhys]

Then run more conservative programs to go slower. In the end you go the same speed as everyone else but spend tons of useless cooling solutions.

Hmm.

Will 2.6 be better than 2.4 or just faster?

[jsyr]

My question is simple.Being faster does not necessarilly mean that is better.I would expect the new kernel to be more versatile and tuned with less powerful platforms in mind(e.g. Strong arm or other embedded system solutions).Computers are nowadays smth like swiss knifes.In the future there will be specific hardware for specific tasks and linux kernel must be there! The need for a project like the GNU Hurd must alert linux kernel developers.

Re:Will 2.6 be better than 2.4 or just faster?

[Anonymous Coward]

Hurd is *not* needed, The hurd hackers just decided to finish what they started. Plus they want to explore the idea of a microkernel.

Even RMS concedes that if Linux was around when they started Hurd they wouldn't have been bothered.

One might also note that they've been hacking at it for longer than Linux has existed; and its still not "made it"

Happy Trails

nameservers?

[phorm]

A prior employer of mine ran 486/33's (and maybe 386's, not sure) as nameservers, and also as webservers too I think. Biggest complaint was compile-time for apache/etc

Logical Volume Manager?

[Hoblin]

I'm looking forward to the network failover and I/O failover features. The Device Manager looks pretty, too. But, when are they going to provide an friggin' LVM? Optimizations are great, but I want features, Damnit!

Re:Logical Volume Manager?

[Gothmolly]

insmod lvm, you putz.

Re:Logical Volume Manager?

[router]

They are working on LVM, and EVMS (userland) is going to work with 2.5/6 like it does for 2.4. If you need more LVM than EVMS, you are crazy. Also, what network failover is new? 2.4 supports channel bonding (read Documentation/networking/bonding.txt) and also supports just about everything you could possibly imagine wrt networking via iptools (read the lartc howto).
Features are the one thing Linux doesn't lack, its got so many tools that, outside of custom applications with extreme requirements, it can basically do anything. Oh, also, can't run freakish proprietary code from monopolists, but what can?
In short, these guys are good, and there is so much under the hood that is really incredible (and has been there for so long....) that outside of vendors supporting their own hardware (like they do for Win) with drivers (GPL please) there is little that is necessary to add to the linux kernel. Not that I don't applaud them trying, or would I stand in the way of a cool hack.
They are even trying to put a generic crash dump capability in that allows userland to dump kernel cores to:
Filesystem of your choice
Serial console
Network device of your choice
Other server
etc etc
Amazing stuff. Just freaking amazing. If you haven't looked over the capabilities of the kernel lately, you owe it to yourself to do so.

andy

A contrary viewpoint

[Magnus Pym]

OK, I have had it up to here with all this FreeBSD worship. After putting up with this for a long time from one of my good friends who happens to be a BSD bigot, I made the mistake of wiping out my Mandrake 8.2 and installing freeBSD on my box.

After a few months of that, I am back back in Mandrake 9.0 with relief and no regrets. Why?

1) I found that, for the things that I do, FreeBSD offered no advantages at all. Performance and stability was no better than Mandrake 8.2. In fact, under heavy loads, my experience is that Linux 2.4.x is much better. (I run lots of octave math simulations and lots of fortran number crunching programs, often several at a time. )

2) For people used to working with linux, there are lots of annoyances to working with FreeBSD. I missed the convenience of RPMS. Many of my favorite programs did not compile properly.

3) When pitch came to shove, my friend had no suggestions as to why the FreeBSD install did not perform as well as linux, except to tell me that I must be mistaken in how well the linux install performed! Duh!

Now, maybe under some circumstances, it is probably true that FreeBSD does outperform linux. But I could not care less. For the work I do (mostly on the desktop, running simulations, running mozilla and xine), linux is demonstrably a better system than FreeBSD.

Magnus.

Re:A contrary viewpoint

[Anonymous Coward]

When push came to shove is what you meant to say. A BSDer would never make an error like that :p

Re:A contrary viewpoint

[Anonymous Coward]

Sorry, for you're an idiot.

In fact, under heavy loads, my experience is that Linux 2.4.x is much better. (I run lots of octave math simulations and lots of fortran number crunching programs, often several at a time. )

And you're point beeing, did you ever try to build/install world with -march=i646 in make.conf?
I guess reading the handbook is too much to ask?
It's like comparing apples and oranges.

I missed the convenience of RPMS.

You can use RPM with FreeBSD, no problem at all.

Many of my favorite programs did not compile properly.

Use pkg_add or the ports, if it ain't in the ports (very unlikely) then just ask on the mailing lists.

Actually, I don't know why I bother, I'm actually glad that I wont have to bother with people who are unable to RTFM.

Re:A contrary viewpoint

[Arandir]

OK, I have had it up to here with all this FreeBSD worship.

Now you know what it feels like to wade through all this Linux sycophantry. Geez, reading Slashdot is like reading about how Linux can do anything including your laundry.

Speaking of laundry, your laundry list of complaints boils down to one item:

1) FreeBSD is different than Linux. Duh! First, optimize the system and applications for your compiler. Second, learn how to use ports instead of assuming that anything not RPMS must be bad. Third, who the hell needs performance during a system install?

jiffies

[thoolihan]

Shouldn't benchmarks of the kernel start including this setting? I'm assuming the default (100) since it's not mentioned, but isn't it worth running test with several different settings? It's affect may vary by load, cpu's, etc...

-T

Already in 2.4

[marm]

But, when are they going to provide an friggin' LVM?

Err... Linux 2.4 has included Sistina's LVM for some time. 2.6 will have a more generalized kernel interface, the Device Mapper, that will allow both version 2 of the Sistina LVM, and the IBM alternative, EVMS, to be built on top. Or at least, that's what Linus seems to have decided on for the moment.

The Device Manager looks pretty, too.

I think that perhaps you are confused. Device manager? And a pretty one, at that? No LVM? Hmm, ok. Maybe you need some spelling help: L-i-n-u-x spells Linux, not Windows 2000. ;)

Re:I'm using P75

[sebol]

I'm using P75 as desktop.

running wmaker, XFree86 4.2
galeon 1.2.x as browser.
pine for mail with fetchmail + sendmail

each time I want to compile stuff,
just "nice blablabla"

[sebol@localhost sebol]$ cat /proc/cpuinfo
processor : 0
vendor_id : GenuineIntel
cpu family : 5
model : 2
model name : Pentium 75 - 200
stepping : 6
cpu MHz : 74.705

It Should Be About Marketing And Money

[duck_prime]

Why does it seem that so many people in the current Linux community *think* that it's about marketing and money, though? *sigh*

Isn't corporate acceptance one of the key points of OSS, especially Linux?

I mean, if it were *just* about technical cleverness, the story would be over. Okay, Linux is clever. Film at eleven.

But there's also the crusade out there to *prove* to "them" that Linux can hack it in the enterprise, that it stacks up against Solaris and Windows NT. Space in server rooms is at a premium, and it's a victory for open source whenever a rack slot gets filled with a linux or bsd box.

There's also the cherished open source community belief that "sharing" and openness comprise a valid business model. In light of that, marketing and dollars are extremely important. The marketplace is a democracy, and every dollar is a vote.

(gasp)

And that's all I have to say about that.

Re:It Should Be About Marketing And Money

[Simon Kongshoj]

'ey mon,
You wrote:

Isn't corporate acceptance one of the key points of OSS, especially Linux?

Open-Source, perhaps. The concept of Open-Source was devised specifically to sell Free Software to the suits. However, Linux is as much Free Software (which is about freedom, and keeping you in control of your computer) as it is about Open Source, so whether it all boils down to corporate acceptance is completely up to the community. Linus didn't write Linux to sell it to corporations, he wrote it because it was fun. I'm not saying that corporate acceptance is intrinsically a bad thing (it isn't), I'm just saying that a lot of people seem to be forgetting about the whole aspect of having fun, when faced with cold cash.

Surprise -- since Free Software is about freedom, I think it's also about freedom from corporations and the marketplace (well-knowing that even suggesting such vile satanism is sure to get me modded far into /dev/null). It's OK that they use the software our community creates -- freedom is all about not taking that right away from anyone, including them -- but we shouldn't be spending our voluntary programming time sucking up to them for money. The vision of a Linux that is completely "de-geekified" that we sometimes hear about may be attractive to suits, but it sure sounds sterile and boring to me.

But there's also the crusade out there to *prove* to "them" that Linux can hack it in the enterprise, that it stacks up against Solaris and Windows NT. Space in server rooms is at a premium, and it's a victory for open source whenever a rack slot gets filled with a linux or bsd box.

Against Microsoft, I completely agree. Microsoft are bent on our destruction, so yeah, we definitely have to fight them. Against Sun, though? I'm not too keen on Sun and their proprietary software, but seriously, one of the things I liked about Linux (and the BSDs) was how it was a free community-driven effort rather than a market contender like the rest. We're not a competitor, we're an alternative. As an old Amiga nut, I was attracted to this aspect because not being a market contender means traditional market methods won't be able to bring us down. It may slow corporate acceptance, but it won't stop the kernel hackers, or the other volunteers working on projects that they work on because they're fun. I think that such a community will be far harder for Microsoft to kill than a bunch of corporate ass-kissers (not that we are corporate ass-kissers, I just think that there is an increasing tendency towards that sad fate), because they're far better at dealing with corporate opposition than community opposition. This is why Linux got them to shit their pants in the first place. Unfortunately, we can already see how all our "selling Linux to suits" efforts have gotten some suits to make the association "Linux == dotcom fad == stupid business models == cash loss".

The marketplace is a democracy, and every dollar is a vote.

The marketplace is, by definition, a plutocracy. Who has more power in the marketplace, you or Bill Gates? If every dollar is a vote, how can it be democratic, considering the very uneven distribution of "votes"?

.......I think I've run out of things to say about this subject too, interesting as it may be. :.)

Re:It Should Be About Marketing And Money

[inode_buddha]

No, the corporate acceptance thing is a relatively new thing, appearing within the last few years. IMHO, the best way to discover the key points of OSS and especially Linux is to ask Linus. After all, it's still his project, regardless of what anyone else does in the macro scale.

M$ Publicity

[Anonymous Coward]

I wonder why Microsoft advertizes in Slashdot.

Sell products? To *NIX users?
Make money? Without selling products?
COnquer new users? Here, in the middle of a Linux Zealots army?
Make us angry, by showing that they exist and throwing in our faces that they have enough money to advertise anywhere?

Mn, the last one seems quite possible.

Re:M$ Publicity

[tuxlove]

I wonder why Microsoft advertizes in Slashdot.

Most /. readers use IE, if I recall the recent topic on the subject. Given that, it's not a stretch to assume that the majority of /. readers use M$ products regularly.

Re:M$ Publicity

[Anonymous Coward]

It must be that they are running SunOs.

Re:M$ Publicity

[Arandir]

So they need adverts touting what they already have?

Irrelevant stats

[tuxlove]

None of these stats seem to cover simulated heavy multiuser/multithread activity. That's what's key as far as I'm concerned. One of the major flaws in Linux today is the scheduling of user processes and file I/O (not sure about networking I/O, but it seems okay from simple observation). There are still severe process/thread starvation problems in the 2.4 kernel which are supposed to have been addressed in 2.5, but I've never seen a really good, real-world performance test to prove it. Until those problems are solved, Linux won't be useful for realtime server work other than web service.

In case you're wondering, no, I'm not a troll. I've done *extensive* testing in this area. So have others, which is why they've been working hard on scheduling.

Grammar check!

[lvdrproject]

Run-on Sentences When Will People Learn?

How the OS's measure up...

[Anonymous Coward]

Linux 2.6: 12-inch steaming love-rod
Linux 2.4: 10-inch muscle of love
Windows XP: 3-inch disappointment, but hey, she's the one who wanted to be a virgin until her wedding night...

Macintosh OS X: Vagina

What do these numbers say?

[greppling]

Not much I'd say. Take "process_load". There are a couple of processes running, some of them called "background load" by Con, the others called "workload".

The workload is a little faster in 2.5, the background load a lot slower. Why should this be better? The kernel compile ("workload") could just be the load YOU are running in background.

These benchmarks are certainly extremely useful for developers, telling them how behaviour in a multi-tasking situation changed. But they don't serve for a "2.5 is so much better/worse than 2.4" (at least not without careful interpretation).

Re:It's faster and rewrites your files for you!

[Anonymous Coward]

I tried to rewrite my /etc/fstab on a 2.4 kernel, and I'll be darned if it didn't undo everything, and create a bunch of new sub-directories in /mnt
So much for swapping a linux install on a slave HDD, to Master, you know, changing your hdb to hda, etc.
Real easy on 2.2 kernel. If that 2.6 kernel gets after your stuff, then you'll just have to install everything where it's gonna go, and leave it be, just like Windows XP!
----------------------
Notice from Slashdot!
Don't listen to anything this fool says! We caught him playing with a copy of WinLinux 2003! This bastard actually REMOVED KDE from it, and replaced that with icewm to run on his cheap, out-of-date crappy computer with not enought space in his Windows partition to install all of WinLinux 2003!
Don't say we didn't warn you...Careful, he's crafty: He'll even claim to have removed all of the "FAILED" messages and replaced them with "OK"' s on his hacked WinLinux!

/r_smp 1

[Anonymous Coward]

Q3 engine based games like RTCW already support this. I doubt Carmack will drop this from Doom III.

Re:/r_smp 1

[Daniel Phillips]

Q3 engine based games like RTCW already support this.

Thanks, dear AC, for pointing that out. In other words, id has already ushered in the era of multi-cpu gaming, and there's no question at all that dual cpus get you more bang for the buck than a single high-end processor, twice as fast. At least this goes for id games, but imho, where goes John, goes the entire 3D game industry in the long run.

So now the interesting question is: when do quad machines hit the sweet spot on the cost/performance curve? I'll go out on a limb and guess "quite soon", that is, 3 years or so from now, and that is entirely due to the fact that AMD has already integrated most of the glue you need for SMP onto the Hammer. Now it's mainly a matter of waiting for quad mainboards to start hitting the overclocker market. Yes, there is an overclocker market, and there are companies serving it.

Quad Hammer for gaming anyone?

P75 as scanner server.

[Anonymous Coward]

I've got four P75 w 32MB RAM as scanner servers for our school. Thanx to the xsane-win32 port the students can access the scanner from any PC in the classroom.

The only drawback is some change in the 0.84 xsane port and forward that makes it impossible for me to run it from the network; it has to run from c:\sane and wants to scan the network every time, so I'm sticking to 0.83.

Talking it to death

[Felinoid]

Every major kernel release has this debate..
It has a bunch of imporvements optomisations etc.
It's got newer faster better ways of doing the same old things.
But... It's doing more than ever before.

It COULD be faster or it could be slower.
Or it could be exactly the same as before.

The real answer is "Wait and see" the rest is just talking...