IBM BIG IRON

[drag]

I am curious, does anyone know exactly how much more powerful a IBM mainframe is compared to a state of the art PC?

At my evening job I basicly sit and obay the commands of my mainframe master. It has me running around fetching him sacraficial backup tapes to abait it's collosal hunger and submitting him a multitudeness of batch jobs by my brethen programmers in a effort to abait the needs of our direct marketing god-kings with the names,adresses and profiles of the great unwashed masses.

Now we run a ibm s/390 series mainframe with a single cpu running jcl batch jobs over vse/esa over vse/power, or something like that. We keep his lordship running at about 95 -100 percent capacity 24 hours a day and 7 days a week exept of course when I screw up a command and everything grinds to a halt for 15-20 mins while I try to figure out what hell I just did. All in all it's very robust and I doubt many PCs would be able to servive that for years on end. Hell if it detects a problem It will even dial automaticly out to IBM to be analyzed by the technitions there, half the time they don't even have to call us back to ask us if we noticed anything went wrong.

But when they told me how much they spent on this beast in one month for liscensing and tech support it about made me want to throw up!

I was curious how many pc's in a linux cluster running vse/370 over hurcules (ibm emulator) would it take for us to equal the power of one IBM mainframe like ours, or maybe a zseries model?
how about without the emulation? I know they run linux in IBM mainframe's VM (<==OS) partitions, how bout the other way around!? (blasphemy I know.. I know, at leat i am tyoing this on a win95 150mhz machine so my master won't find out my treason)

OH crap my master beckons.. I must depart

 

[sinisterDei]

Those numbers might be hard to come up with, I don't think IBM likes to widely distribute performance #s for their mainframe series unless you're a potential buyer, and even then quantifying those numbers up against anything you could come up with yourself would be hard.

Got a question for ya tho, where do you work? Cause I have to play with these things all the time (software side) and it may be my stupid commands making you go recall tapes for me, that is, unless our robot gets it

 

[heliomphalodon]

I had the opportunity to do some testing of a Linux domain running on an IBM S390.
It was running SuSE Linux 2.4.7. but I was never able to determine its clock speed.
My guess is 300MHz.

According to my tests, its integer performance was roughly comparable to a 300MHz Pentium II,
and its floating-point performance was roughly comparable to a 500MHz Pentium III.

Quite unimpressive, but running small compute-intensive programs is not what the S390 is designed for.
It's designed to enslave legions of tape-monkeys!

 

[Buddha Bart]

According to hercules's site, you'd get about 10 MIPS out of a dual p3-1ghz machine.
I don't think anyones going to be using that for backup.

heliomphalodon: what model/family machine did you run those benchmarks on?

 

[capybara]

Originally posted by: Buddha Bart
According to hercules's site, you'd get about 10 MIPS out of a dual p3-1ghz machine.
I don't think anyones going to be using that for backup.

heliomphalodon: what model/family machine did you run those benchmarks on?

how do you convert MIPS to ghz?
im not sure there is a realiable way, bec some instructions take so much longer than others.
let say the average instruction is one-half line long = 40 characters = 320 bits ......?
anyone?

 

[drag]

I think it would be more immportant to know how many mips you can get from a bottom of the line S/390 or a zseries, because that is what I am thinking about.... If hercules (s/370 emulator) could get about ten MIPS from a pentium three, how many could you pull from a s/390 with a single proccessor, we still use s/370 technology the, hell some tape machines I use everyday are older than I AM! (Bacwards compatabilty is one of ibm's strong suits.)

 

[heliomphalodon]

Originally posted by: Buddha Bart
heliomphalodon: what model/family machine did you run those benchmarks on?

I don't know. Access to the domain was provided to my group for testing, with Linux already installed. All we knew was that it was a System/390 domain.

 

[Buddha Bart]

http://www.wilsoncomputer.com/S_390.htm
Looks like their lowest end machine, from the original introduction of the 390 architecture pulled about 11 MIPS.

Some of those G6's break 1000, and the zSeries are faster than those.

bart

 

[AbsolutDealage]

Originally posted by: Buddha Bart
According to hercules's site, you'd get about 10 MIPS out of a dual p3-1ghz machine.

I doubt you would see the same performance from a dual p3 that you have from even an antiquated IBM mainframe. If you are talking MIPS, you forget that the clock cycles per instruction of a mainframe's low-end tasks is much lower than the p3, which has been optimized for more complex tasks. This makes the effective performance of a mainframe much more than that of a desktop (for average mainframe functions).

In addition, I would much rather pay IBM to do service/support on one machine than have to support an army of desktops.

 

[drag]

I know that it is not praticle at this moment to change over to a Linux cluster, mainly because of our antiquated tape and dasd machines, not to mention everybody here are experts here at using that paticular IBM machine. But hey for licsensing for the OS alone costs us like 4500 dollars a month, and that doesn't include the costs of tech support and other stuff we need to support that machine.

Linux/BSD clusters seem to be the wave of the future. And at least IBM has spent quite a bit of money on linux developement if that means anything. Don't forget that the fifth most powerful computer in the world is a linux cluster and that clusters are being used by many leading companies and research places. In fact if a company already has the knowhow a supercomputer-level computing power can be built in-house for a fraction of what it would cost from any other sources, but then again costs are always relitive...

And another thing 50 pcs running various different software configurations and hardware specifications would be a nightmare to contend with. But 50 diskless machines with identical harware and a distruted OS (or disks with a ghosted OS) with a couple dozen machines set up as hotswappable redundent backups can be administrated easily (ie. I could even do it) after it was set up. I fone machine goes down, the backup starts right up assumes the pruduction computer's workload, you get their in the mornig find out one of your PC's went down, replace the bad bit, pop it back on the network and retire the back-up, and take a nap.

 

[drag]

Oh btw, about that linux running on a IBM machine having the capabilities of a 300 mhz PC...
I forgot to say that s/390 Linux just runs in a partition of the main vse/power or VM OS. A partition is just how the IBM devides up the resources of a mainframe. Some partitions are bigger, and they have different priorities. For instance we run the IBM with a vse/power OS and run the older vse/esa OS on top of that to run all the stuff in the computer room, we are what is known as tradition small business IBM center or some equally freakish thing. We have 4 tape reel machines that we run time to time that are older than I am. IBM built its stuff like tanks

IBM has begun a program that you can outsource the running of -say- a websever on linux in one of it's mainframe's partitions. IF you need more power for your server you just call up IBM over the phone and pay more and they will just add more partitions to you own, therfore increasing you server power with just a phone call. (and a increased monthly bill) They can also set up peak times during which you will expect you need greater compututional power, they can also set up burst billing if randomly you need extra power temperarally.

SO I think those benchmarchs are probably from just a fraction of what was realy going on. So unless you know what else the mainframe was doing at the time it is of limited value.

 

[drag]

Hows this for scalability?

 

[RichieZ]

I worked at IBM SVL this summer, and they have 2 of the 64bit Z mainframes. I was testing a new function for the next mainframe version of DB2 and god it got super slow, especially when I would run tests on a distributed database between multiple EC machines.