Slashdot Mirror
Linux On Big Iron
panker writes "eWeek is running an article about a company who converted their IBM mainframe into a Linux email server. "The technical support manager at Winnebago Industries Inc. recently oversaw the deployment of Version 7 of SuSE Linux AG's Linux operating system on an IBM zSeries mainframe to run his company's e-mail server supporting 700 users." "
IBM has a Virtual Machine OS, that allows you to run multiple OS's on a mainframe. You can run Linux (or even multiple instances of Linux) and still run your legacy apps under OS/390.
[Insert pithy quote here]
That's possible too, but head on over to Matt Simerson's FreeBSD Toaster. I'm SURE that could easily support a few thousand users in a clustered environment (NFS & Mysql)
Need more users? Add another box.
"I can't give you a brain, so I'll give you a diploma" - The Great Oz (blatently stolen sig)
more like shipping horsepower; it's a way of coming up with a single metric that normalizes the range of machines/configurations the company sells -- kind of like a capitalization-weighted stock index; the big boxes count for more, which they wouldn't if you just went by units shipped.
"It's not a bear, it's a hamster. A really, really large hamster."
Imagine that IBM sold three models, with the following MIPS capacity:
Model A: 10 MIPS
Model B (Linux): 100 MIPS
Model C: 1000 MIPS
And they sold the following amounts:
Model A: 5000 units (total of 5000*10=50000 MIPS)
Model B: 500 units (total of 500*100=50000 MIPS)
Model C: 50 units (total of 50*1000=50000 MIPS)
Don't you think a statement like "33% of our sales (by MIPS capacity) was configured for Linux" is a little more informative (and accurate) than "9% of our sales were Linux servers"?
Big Iron is generally measured in MIPS, (been that way for over twenty years). So what the Big Blue Spokesperson is saying is:
"In the fourth quarter of last year, eleven per cent of the total computer power we shipped was tunning Linux."
Now that might mean that they shipped a total of 100 Mainframes (Really Big Boxes) of various models. They added up the MIPS of all of them and came up with some number of total MIPS -lets say 100,000. Of that 100 mainframes, thrity of them (relatively low end) totalling 11,000 MIPS were configured with Linux.
I'll agree, it seems kinda dumb, it would be like Ford reporting sales based on the total horsepower of all the engines in all the cars and trucks they sold, and then giving the percentage of them "configured" for diesel.
Maybe an UBMer would care to explain why it makes sense?
You either believe in rational thought or you don't
When you buy a mainframe from IBM you pay for the processing you want. They ship you a complete system (anywhere between 4 and 16 processors I believe?) and turn on (and charge you for) only the number you want. Some of the processors can be configured to only run Linux, and that is what they mean by "mips" being configured to run Linux.
The reason for configuring the processors to only run linux is that many mainframe software vendors charge based on the number of MIPS your machine runs at. If I am buying an MVS application I do not want my Linux processors in the mainframe to affect the price I am paying.
Finkployd
...according to this arcitle on IBM's Website.
Perhaps it's just me, but that makes no sense whatsoever. How do you ship a measure of speed? "Shipping" millions of instructions per second seems to me to be the same as "shipping" miles per hour.
;-)
It's just you.
In the mainframe world, where virtualized hardware is the norm, systems are sold by the MIPS. In other realms of computing we talk about a 32-processor or a 128-processor system (or, if you buy Sun, a 106-processor system, for some reason). In mainframe land, you talk about a 12 MIPS system or a 45 MIPS system or whatever.
It makes a lot of sense, too, when you think about the fact that a four processor system, mainframe or otherwise, from five years ago is probably less powerful than a one-processor system now. It sort of normalizes it if you talk about system capacity in MIPS rather than in terms of "x processors of type y at z megahertz."
Don't compare the practice to shipping "miles per hour," but rather to horsepower. General Motors could, if they wanted to, say that they shipped umpteen million horsepower worth of engines last year, and it wouldn't be that hard to understand. It's just a different way of counting.
I'll agree, it seems kinda dumb, it would be like Ford reporting sales based on the total horsepower of all the engines in all the cars and trucks they sold, and then giving the percentage of them "configured" for diesel.
:)
IBM charges you for your mainframe (and support contract) based on MIPS. The software you run in MVS (in most cases) is priced based on the number of MIPS you run in your mainframe shop. Ford doesn't charge for their cars based on horsepower
It makes sense because it lets you know that 11% of all mainframe computing power sold in the 4th quarter of last year is running Linux. IBM doesn't sell "boxes" per se, they sell MIPS.
Finkployd
On their minis (AS/400) they are known as Logical Partitions, which allow the machines to run Linux, Notes servers, and various versions of OS400, and rumored to eventually include other OSes...
Many times easier to support one machine that multiples... and its easier to execute a backup machine for it as well.
* Winners compare their achievements to their goals, losers compare theirs to that of others.
But if you read the article you see that they don't want to support more servers, they want to support less, i.e. not buying anymore Intel servers, which are like so many cats. The only downside I see is a single point of failure, the zSeries goes 'poot!' and the staff takes the afternoon off.
But adding that function to an existing piece of hardware does keep support costs down, and as they've noted, pay once to get their mail running on there, as opposed to paying Microsoft for Exchange, year in, year out, well, it looks smarter, doesn't it?
Last, but not least, if they decide to move it off the mainframe later, hey, they should be able to migrate it with little pain, since the OS runs on just about anything.
'Bago 2005: "Tell eWeek we just moved the entire mail server to a hacked TiVo, will you, don't forget to mention it does voice and video email, too."
A feeling of having made the same mistake before: Deja Foobar
My father now works for IBM (bought and outsourced to them) so he gets all the newsletters and the such. Last year IBM had something akin to 20x mainframe sales after the linux initiative. It pretty much saved the department.
The only problem I've seen is most current admins are used to/learned linux on little dinky spare desktop machines. 'Mainframe' carries a big scary connotation. The name itself intimidates, like a *nix prompt scares most MCSE's.
Plus most bosses 'know' that mainframes are *so* 1970's...
They didn't do anything with the old apps. That's what I meant by "logical partition". You essentially create an address space that thinks it's its own machine, and can run a different OS.
The old stuff still runs without modification, albeit with slightly less resources.
Garg
Garg
Alumnus, Xavier's School for Gifted Youngsters
Actually, the mainframe can be sliced into logical partitions, LPAR. My guess is that they already had a mainframe which they use with one of the mainframe OS's to run their accounting, inventory, or manufacturing. They allocated a small LPAR for this, put Linux and the mail server software on it and voila, no more exchange. The software they use (Insight Server from Bynari) allows running exchange type capabilities from outlook, so they replaced MS Exchange (which for 700 accounts is something like $50,000-$75,000 including OS, mail server, hardware for at least 2 intel boxex, CAL for both OS and Exchange), maybe more and not including Sysadmin, with this.
I note that there is quite a few peoples that do not know how things are done on a mainframe. I will TRY and clarify. If you have an average box, say 400 mips, there will be a couple of things happening on this box. Starting of with the OS, OS/390 (or z/OS, VSE, VM). Then a network server VTAM and TCP/IP, a Security server, the some DB's DB2, IMS/DB(or some vendor DB). Then transaction servers CICS or/and IMS serving 4000+ concurrent online users logged on to CICS or IMS. Then you may have oh lets say 500+ programmers, system programmers, DBA, Administrators etc, logged on to the OS "shell" TSO doing programming, compiling, admin, editing and in general doing what these kinda people do, maintaining the monster. The there may be a MQSeries or two running handling client connections and messaging applications and client (pc's, server's *nix's) connections to DB2.. (i'm touching the surface here!) So in general there is a little more happening on a mainframe than on you average wintel box. So in order to separate the production and development, and test environments you can go and partition this one little box up into three logical partitions, called LPAR's in dinosaur speak. Each LPAR can be IPL'd (BOOT in dino speak) without affecting any one of the other LPAR's. So now you have one IBM 2064 400MIP box, 3 LPAR'S and still have room to breathe. Add another LPAR, install VM, Load LINUX/390 and reboot the LINUX LPAR. But still only one Linux server on the box? But we have 250 NT boxes to replace? Not to Worry!! under VM on your brand new LPAR on your mainframe box you can begin to start Linux images at will. So in the end we have 1 Box, 3 OS/390 Mainframe partitions, 1 VM partition with 250 Linux servers running. But now the 400 mips is kinda running out of steam. Call up IBM and they will come and add another cpu. And if you were planning ahead the IBM 2064 should have an idle couple of CPU's under the hood not being used. So you call IBM, they give you the code and you go to the master console and issue a "very cpu online" command and off you go. No IPL required, no downtime. Do that on your duel XEON Intel with Windows... any kind of Windows. I'm not going to go into Parallel Sysplex'ing, syscon's and CICSPlexing etc.