Slashdot Mirror
Exchange vs. Linux/390 Comparison
eclarkso writes: " The Consulting Times has done a quite even-handed study of the TCO for each platform in a fairly large (5000+) enterprise environment. The article is as much a commentary on the mainframe architecture as it is on Exchange vs. Linux groupware."
Some of the academic folks I know have had a bit of trouble installing Linux/390 (<----- ibm's linux/390 developer page), but linux390.marist.edu/ has a decent manual they've found helpful.
Of course, it'll long be obsolete before I ever get my hands on one of these beasts. *sigh*
Did we count the difference in functionality? Exchange vs. what on Linux?
The mainframe may be back -- but make no mistake it is still the domain of the priesthood. The priesthood that the server architecture was to break up. Do Linux users really want that? A handful of techs who are well paid (the business people are cheering) but no need for the thousands of SA's and small shops can just buy time on a 390.
Not entirely...
If this is a BIG IBM mainframe then it will take
more floor space than a single rack of twin processor CPUs.
I assume that VM programmers are in short enough supply that paying one $90k p.a. is reasonable. It's not far from what I get paid as a Unix SA.
And the hardware is WAY more expensive.
The second set of figures shows how much less you will be paying if you already have an IBM mainframe (for some other purpose) that you can use for a Linux partition (virtual machine, whatever you want to call it) compared to bringing in an NT server farm with exchange. You've already paid for the hardware, the floor space, the support staff. You're just pushing your hardware a little harder.
Does it make sense now?
Z.
-- Under/Overrated is meta-moderation, and therefore is Redundant.
Can anyone help on this one?
Answer: You're lying. You're talking out of your ass based on specs and dead reckonin'. They did the numbers, you did the empty speculation, and guess what? Reality wins, hands down.
by Mike Buddha -- Someday the mountain might get him, but the law never will.
I jumped at IBMs offer for developers to try out Linux running on a z/Architecure thingy. My experience so far has been pleasant and "boring" (in a positive way). Pleasant because everything ported easily. Boring because I was expecting challenging porting problems. Everything worked.
Now for the article about TCO and stuff... I believe that it is correct. For small installations use cheap hardware to bring down initial costs. But don't be afraid of mainframes when you business grows - they are not that different.
What email/groupware software are they using on this Linux/390 machine? Is it some port of Lotus Domino server? I am only aware of Domino running natively on S/390 and on Intel x86 Linux, not on 390 mainframe Linux. Also bear in ming that there is no antivirus vendor supporting Domino on any Linux platform so that makes Domino on any Linux platform rather useless, doesn't it?
"For the sake of simplicity, certain items such as depreciation and management overhead were excluded from the comparisons."
It's kind of interesting, since management overhead is widely regarded as the main reason why people prefer Windoze systems to Linux systems. People believe that it costs less money to perform essential administration tasks in Windows than it does in Linux.
I'm not stating that the costs actually are lower, but it's not a terribly informative article if they're going to eliminate that important bit of information.
Why do you need VM programmers? The port is already done, the logic for running Linux as a guest OS is there, and it's stable. Henceforth you should be coding on the Linux level, not the VM level.
Has anyone tested opensource freeware for 25K users? You could save money on different designs, but this really isnt the meat of the article.
I know it can be done cheaper, we have designed email/groupware for millions of subscribers cheaper than that, all are webbased with oracle/ldap on sun equipment, with network load balancers.
I like how Exchange looks like a cheap solution, untill you grow past your user base, then costs sky rocket.
Now that's an interesting point! My experience has been that when overloaded...
1) mainframes and real Unix servers (Sun, HP, etc.) slow down instead of crashing.
2) Linux (and NT) crashes hard.
So the question is, does the OS crash on a given platform because of the hardware, the software, or a combination of the two? What will Linux on a Mainframe do when hit with an enormous load?
Unless the kernel has been rewritten extensively to deal with the hardware, I suspect it would crash just as effectively on an S/390 as on a stack of Pentiums. I'd love to find out for sure, though.
"People who do stupid things with hazardous materials often die." -- Jim Davidson on alt.folklore.urban
If I did understand it right, then 70% of the TCO was always the support personal cost. So if there is no need for support personal for IFL, its clear
that it rocks. The thing that I didn't read in the article is WHY it does not need to support.
If programs would be read like poetry, most programmers would be Vogons.
There's no way Exchange2K could handle 50K users on a single box.
First, you've never obviously worked with Fibre Channel on the kind of scale that 50K users would require (ie. a big EMC box)...that many users pounding the same box will easily chew up 50% or more of your CPU power on I/O alone. Fibre is fast, but it is so fast that it can easily swamp Xeon CPU's. I know, because I did the benchmarking at my company.
Second, connection limitations in Win2K and Exchange alone mean that you are running very close to the theoretical maximum the OS supports..not a good idea.
Third, running that many users off of a single box is suicide. And if you've ever watched Exchange2K failover on a Win2K cluster, you'd know that it can take several minutes for everything to come up on the second node, if you've got a lot of users.
Finally, a 100GB array for 50K users results in a 2 megabyte mailbox..that's freaking ridiculous!
In short, you're either running a 50-user shop, or you have no idea what you are talking about.
We tested both server platforms using a custom test suite that simulated a large population of simulated POP/SMTP users.
*POOF* There goes that argument!
How about some data that reflects how people actually use Exchange (through Outlook RPC). I also can't see what the client settings are -- maybe they're only pulling new mail once an hour or so.
(I've seen MS put out Exchange "scalablity" numbers using POP3 before. Easy way to beat up on Notes or Groupwise, but POP3's the game, a Unix box or that mainframe should be able to handily kick it's ass.)
Business. Numbers. Money. People. Computer World.
Somehow those numbers look pretty high - especially if you look at the solutions other companies run...
3 6xXeon systems 2 to 1 failover $80k
1 Linux retail box $75
2 Admins @ 75k/year $150k
-----
$230.075
Well, dont know but somehow this whole linux on mainframe seems like overkill for me - especially since the mainframe CPU's arent all that impressive and the linux vm's dont profit all that much of the datatransfer rates a mainframe offers...
Um, don't get too excited about this report:
The custom test was designed by eTesting Labs to simulate from 33,320 to 83,300 POP/SMTP users that checked their mail every 60 minutes and sent a single 10K byte message to three recipients every 60 minutes.
Honestly, if you are using your Exchange server as a POP/SMTP server only you are wasting your money. Exchange is groupware, you do not use it as a POP/SMTP server. Save your money and just run sendmail on Linux, BSD or Solaris. Exchange is for calendering, scheduling, messaging, etc. This report is pretty much worthless.
Q.
Comment removed based on user account deletion
aside form the fact that the first IBM system he first describes is vastly overpowered, and a rediculous 'solution' to supporting 5000 users.. It doesn't matter if it can scale to 50,000; that's not what it's there for.
Why compare it on big iron? Why not compare it solely on the same hardware?
I can support 50,000 users doing all kinds of neat things on the same hardware, running linux, for a LOT less money.
Notice the Exchange licensing costs? a quarter million bucks?
Keep in mind; most companies do NOT use exchange for what it is good at.. they use it for pure email, though they may purchase it thinking they will use all the groupware features.
I am a MCSE who has a love/hate relationship with MS and their products. I really like Exchange.
A lot of exchange shops do stick to a 300-350 user limit per box for Exchange 5.5, but that is with the following conditions:
No real company has 10 meg mailbox limits
Until the current generation of tape backup (ultrium, superdlt) came out, having a mail database (priv.edb , the "priv in exchange speak) muuh bigger than 20 gig really alarmed people due to SLA's for restoration of service in a server corruption/failure scenario.
So, if you assume for their scenario that they were running E5.5, I would have put at least 1000 mailboxes per server, probably 1500, allowing me to max out at 15 gig priv. This would cut down the hardware costs considerably.
With exchange 2000, clustering is a lot more viable, and e2k also allows a lot more (up to 16, instead of 1) private stores (databases of email) per server. MS has had some issues with MAPI clients and clusters , so I am really hesitant to say how many more users I would put per box.
Overall though, I think its clear that if you have tons of users, linux on big iron can make a ton of sense. Comparing qmail/sendmail to exchange is somewhat unrealist on a features standpoint, but for the major league web email providers, big iron must be worth looking into.
I really think the 10 meg per user mail limit somewhat discredits the whole analysis though. Sounds way more like webmail than corporate mail
ostiguy
My experience as a user of Exchange is that if you let the administrator do a traditional Microsoft Office closed-system implementation, you're forcing all of your users into using an appallingly bad piece of software which leads to horrendous support problems down the road. It's not just the Virus Of The Week problem - Outlook Mail, while much much better than some of the previous MSMail products, fundamentally doesn't get it, and it keeps the user's mail in one big honking file that's increasingly fragile and bloated, and has an undocumented and unrepairable format - if it croaks beyond your client program's self-repair capabilities, you're hosed. It also Encourages Users To Mail Around Attached MSWord Documents or several other proprietary formats instead of just sending the message as real plaintext - leads to extra work for the reader (and usually sender), and bloats mail substantially, so your system has to carry a factor of 3-10 more traffic.
Exchange also encourages the users to send mail around with Internal Email Addresses - messages appear to come from "Joe User, Marketing" instead of "juser@foo.com", which looks pretty but fails badly whenever mail gets forwarded out of the system - if you send mail to Joe, Jane, and Fred@customer.com, Fred can reply to you@foo.com, but doesn't have a way to reply to "Joe User, Marketing" or whatever Jane's fictitious title is.
It's not like Sendmail doesn't have a long history of evil on its own, or like you can't build Turing Machines out of sendmail.cf files. But at least it's open, documented, and transparent, and runs on real operating systems.
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
"Fred can reply to you@foo.com, but doesn't have a way to reply to "Joe User, Marketing" or whatever Jane's fictitious title is. "
Huh?
Exchange automatically does the conversion when it goes out the SMTP layer.
Folks,
.. In the corporate world, you have to be able to do things such as "recover" a significant (L)user deleted email. If the CEO says "whoops, I poo-poo canned it accidentally", you're expected to fix the situation..
You're missing the entire point of deploying a messaging system in a corporate environment. This is what messes Linux up. It's nice that you can run SendMail, popD and whatever on the big hardware.. but.. for my corporate end users, this isn't adequate.
Here are my criteria, sorted in no particular order, for a system that I would be happy to deploy to my 700+ users:
1) Reliable: No loss of data (no PC storage, backups are centralized). [admittedly, tough to maintain with exchange, in the field]
2) Useability: (l)Users can find their info quickly and easily. (search via header, sender, date, text in body, text in attachments, etc..)
3) Manageable costs associated with the above two criteria. I'm not claiming $0 cost -- but predictable and manageable costs.
That's it. Exchange rules at meeting those criteria. I don't want to backup 700+ PC's -- I don't run an ISP!
Which is quite common, for the market that Linux is "trying" to target -- except that most implementers assume there is a *nice* SLA in place.. the small/medium size market is not ready for the lack of end-user features that are present in the *VAST* majority of the distributions.
gimmie M$ Small Business Server vs. a Linux/POP3/IMAP solution and I only have to wait until the first end-user "OOPSIE" as a sysadmin, before I toss linux out the window..
Cheers,
Scoots.
Regarding #1:
How good an Exchange admin are you if you don't know about Single Instance Store in Exchange? Priv.edb doesn't hold a separate copy of that attachment for every user. It's kept in the database and is only referred to once for everyone in the distribution list.
So to keep it simple for you:
If your CEO (who you think is an idiot but somehow makes more money than you) sends his PowerPoint attachment to 20 people internally (and assuming those 20 users keep their mailboxes on the same Exchange server), there will only be one copy of that PowerPoint app in priv.edb.
Got it?