• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Linux on a IBM dinosaur...

D

drag

Elite Member
This type of linux server can only support up to 32 proccessors...
Do you think that's a lot?
And it is limited to 256 memory!! jeez, thats pretty bad.. Well maybe not since is 256GB not MB...

🙂

I want one. ANd to that means I have started a little charity, Go ahead and mail your paychecks to:
DragNeedsAnewComputer@please.com

I accept visa/mastercard credit card numbers (please include a jpeg of your signiture/experation date, home address, mothers maiden name, your birthdate, social security number and anything else you can think of)
Paypall is acceptable too, but I'd rather have a box of signed blank checks instead. E-mail me for details...

 
If you want one that bad, then go post in off topic asking every member to send you 5 bucks 😀
 
Oh, I don't know that...
I realy don't know a lot about IBM's stuff, but...

I would assume that 32 dual-core g8's could outperform a power4 machine.. or something.

I thought that this was the high-end of high-end mainframes that IBM was producing?
 
Hmm... Hard to say. Usually stuff like that is a custom job and they don't usually give hard and fast rules, but I would suppose that this guy would easily cost well over a million dollars, possibly for a entry level job with just a couple proccessors..

But that wouldn't be the main cost, just running it would be a big deal. With electrical costs and service contracts and the such you'd be looking at possibly 7-15 grand a month to run it.

Now I am pulling this pretty much out of my but, but I know that with my job's mainframe will cost them 4-5 grand a month in contracts with IBM. But this is just a little single proccessor s/390 job that's about 5-7 years old. I wouldn't be supprised if somebody told me a state of the art IBM mainframe would cost ten times that much a month.

But with that you get 24-7 monitoring and quick response if anything goes wrong. If the mainframe senses a disturbance in the force it will automaticly dial home to the main facilities in texas (or somewhere) and give a detailed report to the technitions, who will examine it and if they think it could possibly be a problem they will call you asking: "Hey, this is IBM, have you been noticing any thing funny with your computer lately?".

I bout s**t my pants one night when I heard the mainframe's modem kick in! I was all looken at it trying to figure out were the damn telephone line came out.

But what is just mortal money worth when compared to the ability to run a couple of thousand completely independent Linux "servers" on a single machine?
 
Yeah, most companies don't buy a system that big outright. They would lease it through an IT financier, much like a fancy car or a plane.

The neat thing about these big zSeries mainframes is that many of them come with extra processors that are disabled on arrival. If the customer ever needs the extra processing capacity, all they need to do is call IBM. Their service technicians can activate the extra processors remotely, and then bill the customer for the extra per processor cost.
 
couple notes:

a 32-way power4 pSeries is drastically faster than this at cpu-intensive workloads.

The cpu's are dual-core, but not for performance. What happens is they both do the same work, and compare answers. If there is a difference, that CP is shutdown and another brought online. All the instructions and data in cache are reassigned, and the process never even hangs. Think of it as RAID-1 for cpu's.

despite dave's test that ran several thousands of linuxVM's at once, you could only get about 30 - 60 linuxVM's doing any actual work on one of these. Thats 60 if you've got stuff like dhcp, dns, static-content-webservers, etc. 30 (or less) if you bring websphere, db2 or domino into the picture.

A single instance of Suse Linux Enterprise Server 8.0 for a z/Series machine costs $15,000. They get around the GPL by calling it a support contract, but you can't get the installer/configuration software (not-gpl'd) or binaries, or even a CD (source or binaries) without said contract.

Redhat's was a little more reasonable IIRC, but I don't have the numbers here at home.

The debian port is still a work in progress (and is in a really rough spot because IBM refuses to have anything to do with debian), and who even knew turbolinux was still in business? 🙂

Truth be told, *nix and s/390 don't go well together. The memory management, scheduling, multi-tasking, and fundamentally interactive nature of most *nix workloads are very much at odds with scheduled/batched and extremely non-interactive workloads the s/390 is so tuned for. Thats why IBM never ported AIX itself.

back to my 3270 terminal
 
It's not a s/390. Its a z series 990. IBM's brandnew top of the line mainframe they just released.

s/390's are the bottom-end of the mainframes, the OS that is used in the s/390 was never realy designed with multiple virtual machines. It is for the mom-and-pop level of IBM houses were they do not want to pay for the extra fuctionality. At my work they use it to run the s/370 era JCL language and tape machine hardware. IBM is well known for supporting 100% backware compatability. They do this by simply running the older OS inside the new one with no performance penatly, out of this developed the multiple virtual machines that are specificly targeted at replacing a network of unix servers by a single box. The version of linux they've modified is specificly designed to run in the z series's partitions.

I understand the argument that the batch-type enviroment is not well suited to the varing demand loads that are common in PC-type hardware. This is due to the fact that people who generally run mainframes know everything about the machine, they do everything in their power to streamline the proccess to suit different enviroments. These enviroments are generally used for "batch" style programming and what they do to there machines is not suited to the normal pc enviroment. They aim for 100% workload at 100% of the time without bogging the machine down, but this doesn't mean that it can't be used for other purposes efficiently and successfully. Saying that mainframes are suited only for batch style enviroments is like saying IBM doesn't know how to use their own computers and are worthless at designing mainframe architectures. And that PC's are not suitable for high cpu intensive tasks like compiling a kernel, or rendering a 3-d image because they spend most of their time idle and can't be stressed.

I know I am not a expert or anything at IBM crap, I personally would think that clusters offer a better solution, but saying that IBM's sceme for linux time-sharing is not a good idea because linux doesn't go well with a ten year old mainframe architecture and operating system is bogus.
The more I am around the s/390 the more i wonder if a 3-4 state-of-the art PC's running parrallel in a cluster could run circles around it in terms of computational perfromance. Probably why IBM is offering s/390 linux is just because they can, and lots of people (well not realy LOTS) realy wouldn't mind running a e-mail sever or two from a partition while they go about their daily batch-running business. I know that the rest of the people at my job are traditional Redhat freaks and have been running it for various things since it's inception, (even though the Big Boss incharge of the computers and half the programming staff is still using OS/2 for his desktop he has a nice big redhat sticker up on his office window.) Maybe it stems from a subconious level of resentment from MS becomming bigger controller of the PC market that IBM started, but loss because they were being to big of a a****les to suit everybody else.

IBM not sharing it's code though is nothing new, it's a big mean bastard of a company. But I can understand a little bit of it's reluctance. The lawsuit from SCO is a indication of what IBM has to look forward to. It's has imparted some code from it's AIX os into linux, And systemV code being gpl'd is a weird turnaround in the unix wourld, but SCO is very pissed from it because it's beginning to realize that it's sceme of having 2 levels of operating system, a high-end unix for the big guns and a low-end linux for everybody else is going to be threatened by IBM using the "low-end" linux combined with select parts if SCO's own code they sold to IBM for the creation of AIX that the linux will end up being a bigger badder OS then the SCO's own Open Unix crud. Once IBM gets around to squishing SCO like a bug (which is what some people think SCO managment wants.. ie good opt-out packages from IBM for themselves and their stockholders when IBM consumes them) , I think that code will end up being a bit more forthcoming. But that's just wild speculation. From the history of system V code and it's many creators including Microsoft I could just imagine the legal nightmares that that would create to just give it to hardcore gpl OS's like debian...
 
It's not a s/390. Its a z series
Click to expand...
Sorry for the confusion, I use the terms interchangeably. Much like saying "x86" or "IA-32".
s/390's are the bottom-end of the mainframes
Click to expand...
Nope, its just a name change for a new generation. IBM decided that s/390, as/400, and rs/6000 didn't represent a unified brand, so they united them under the "eServer" brand, and gave them each a letter, so
s/390 -> zSeries
as/400 -> iSeries
rs/6000 -> pSeries
intel -> xSeries
out of this developed the multiple virtual machines...
Click to expand...
Yep, that plus the rather arcane licensing system
that are specificly targeted at replacing a network of unix servers by a single box
Click to expand...
Nope, not even close. Mainframes are designed for a completely different type of workload. Look at pSeries/AIX... IBM wasn't trying to eat its own marketshare (SP2's get just as expensive as frame's) its selling two different products for two different markets.
Saying that mainframes are suited only for batch style enviroments is like saying IBM doesn't know how to use their own computers
Click to expand...
Their marketing department sure doesnt. If your read the technical literature from the server group however they strongly point out the performance disadvantages of linuxVMs. They make it very clear you will sacrifice performance for managability.
And that PC's are not suitable for high cpu intensive tasks like compiling a kernel, or rendering a 3-d image because they spend most of their time idle and can't be stressed.
Click to expand...
No, PC's are much better suited for high cpu tasks. The CPU's in mainframes are suprisingly weak. I could give you the GHz but we both know that doesnt mean much. The 390 architecture makes massive sacrifices for backwards compatability and reliability that make it a very CPU weak platform. People make fun of x86 for being overly backwards compatible, but they've got nothing on the 40 years of code a zSeries can run.
saying that IBM's sceme for linux time-sharing is not a good idea because linux doesn't go well with a ten year old mainframe architecture and operating system is bogus.
Click to expand...
1.) 10? try 40
2.) The OS actually greatly helps. VM goes to great lenghts to make up for the awkward match of linux and mainframes
3.) Its not even slightly bogus to say that a scheduler, memory manager, and device manager (kernel) need to be optimized for the architecture they run on. Thats what they're for!!!!

Sorry, I didn't want to turn this into a debate, I was just looking to add some facts to the thread with the first post.
If anyone's interested in more info on mainframes and linux-on-mainframes, here's some links:

IBM Technical info and oficial stuff
http://www-1.ibm.com/servers/eserver/zseries/os/linux/
http://www.redbooks.ibm.com/redpapers/pdfs/redp0222.pdf
http://publib-b.boulder.ibm.com/Redbooks.nsf/RedbookAbstracts/sg246863.html?Open
http://publib-b.boulder.ibm.com/Redbooks.nsf/9445fa5b416f6e32852569ae006bb65f/e2ba7b0bdabbfdde85256cb8005cb4f9?OpenDocument&Highlight=0,990 (you're lucky, they trimmed this down over 100 pages since I read the z/900 one)

A very detailed explanation of linux/390's ups and downs
http://www.linuxworld.com/site-stories/2002/0416.mainframelinux.html
http://www.linuxworld.com/site-stories/2002/0506.mainframelinux.html
http://www.linuxworld.com/site-stories/2002/0522.mainframelinux.html

Dave's response to the above
http://www.linuxworld.com/site-stories/2002/0717.boyes.html

General linux390 news
http://www.linuxvm.org/

The Linux390 Mailing list
http://vm.marist.edu/htbin/wlvindex?linux-390
NOTE: Its not your connection or browser, that is quite possibly the slowest loading page on the internet. Don't even try using the search feature, the http connection will time out. One of the guys on the list keeps an archive on his site that serves respectably, I don't have the URL handy.
(oh and FYI the slow page is run of a z/900, though the problem is with the retarded staff, webserver, and mailinglist software)

So anyway, read all that (hah!), follow the mailing list for 6 months, and admin a couple of production linux390 systems for a few months, then come back and read this thread. You'll find you agree with me a lot more.

(longest post ever)
 
Originally posted by: drag
Hmm... Hard to say. Usually stuff like that is a custom job and they don't usually give hard and fast rules, but I would suppose that this guy would easily cost well over a million dollars, possibly for a entry level job with just a couple proccessors..

But that wouldn't be the main cost, just running it would be a big deal. With electrical costs and service contracts and the such you'd be looking at possibly 7-15 grand a month to run it.

Now I am pulling this pretty much out of my but, but I know that with my job's mainframe will cost them 4-5 grand a month in contracts with IBM. But this is just a little single proccessor s/390 job that's about 5-7 years old. I wouldn't be supprised if somebody told me a state of the art IBM mainframe would cost ten times that much a month.

But with that you get 24-7 monitoring and quick response if anything goes wrong. If the mainframe senses a disturbance in the force it will automaticly dial home to the main facilities in texas (or somewhere) and give a detailed report to the technitions, who will examine it and if they think it could possibly be a problem they will call you asking: "Hey, this is IBM, have you been noticing any thing funny with your computer lately?".

I bout s**t my pants one night when I heard the mainframe's modem kick in! I was all looken at it trying to figure out were the damn telephone line came out.

But what is just mortal money worth when compared to the ability to run a couple of thousand completely independent Linux "servers" on a single machine?
Click to expand...

That's a lot of $$$! I guess i'll have to stick to single CPU machines 😀!
 
Originally posted by: Buddha Bart
No, PC's are much better suited for high cpu tasks. The CPU's in mainframes are suprisingly weak. I could give you the GHz but we both know that doesnt mean much. The 390 architecture makes massive sacrifices for backwards compatability and reliability that make it a very CPU weak platform. People make fun of x86 for being overly backwards compatible, but they've got nothing on the 40 years of code a zSeries can run.
Click to expand...

I'm a bit in the dark here, but what do they do to compensate for this? Sheer number of cpus? Memory size/bandwidth?
 
Originally posted by: BingBongWongFooey
Originally posted by: Buddha Bart
No, PC's are much better suited for high cpu tasks. The CPU's in mainframes are suprisingly weak. I could give you the GHz but we both know that doesnt mean much. The 390 architecture makes massive sacrifices for backwards compatability and reliability that make it a very CPU weak platform. People make fun of x86 for being overly backwards compatible, but they've got nothing on the 40 years of code a zSeries can run.
Click to expand...

I'm a bit in the dark here, but what do they do to compensate for this? Sheer number of cpus? Memory size/bandwidth?
Click to expand...

My guess is reliability.
 
So anyway, read all that (hah!), follow the mailing list for 6 months, and admin a couple of production linux390 systems for a few months, then come back and read this thread. You'll find you agree with me a lot more.

(longest post ever)
Click to expand...



Heh, i'll be sure to do that. THis is a subject that I am interested in very much.
so your saying that this is just a s/390 series that has a different marketing name?


 
Originally posted by: BingBongWongFooey
Originally posted by: Buddha Bart
No, PC's are much better suited for high cpu tasks. The CPU's in mainframes are suprisingly weak. I could give you the GHz but we both know that doesnt mean much. The 390 architecture makes massive sacrifices for backwards compatability and reliability that make it a very CPU weak platform. People make fun of x86 for being overly backwards compatible, but they've got nothing on the 40 years of code a zSeries can run.
Click to expand...

I'm a bit in the dark here, but what do they do to compensate for this? Sheer number of cpus? Memory size/bandwidth?
Click to expand...

R.A.S.

But they do have massive I/O bandwidth as well, theoretically bigiron from Sun should be able to compete, but I remember reading articles that had a zSeries easily beating a SunFire 15K in real world sustained IO.

But really, these things are around cause people want backwards compatibility, we have a few customers at work that run s390/Z stuff, which always seems to cause problems for us, since we don't have any personel who know jack about s390/Z, so we can never help them out when they run into trouble. 🙂
 
Originally posted by: BingBongWongFooey
I'm a bit in the dark here, but what do they do to compensate for this? Sheer number of cpus? Memory size/bandwidth?
Click to expand...
As sunner said, RAS and I/O are the two most important things to a mainframe. For RAS, nearly everything in a mainframe falls into one of three catagories.
1.) Hot swapable
2.) "already there"
3.) built in redundant hardware. (like the dual-core cpu's I explained above)

by "already there" I mean things like the number of CPU's and the amount of ram. For instance, all z/990's will ship with 32 cpu's. Its just a matter of how many you've licensed to "turn on". Similarly, they all ship with much more ram than the customer orders, and if more is needed it can just be "turned on". Because of this more CPU's and RAM can be added to a machine with almost zero down time. Those two are however in the "disruptive upgrade" catagory as they require a reboot.

As for I/O, all devices are broken off into dedicated channels with their own processor (CHPID) to handle interupts and DMA. Each channel, and the memory subsytem have massive bandwith (check the redbooks for exact numbers). Also, the CPU's have 512K of L1 cache (256 data/256 instruction), and half of the cpu's on each frame share 32MB of L2 cache. The whole goal of mainframe programming is to get your loop/batch-job into that 256k of l1-instruction cache, and leave the L1-data, L2, memory, and channel I/0 up to feeding that loop as fast as possible. Remember "C" is a newfangled language to mainframes, PL/1 and COBOL don't have anywhere near the program size of modern languages, and s/390 is an extremely CISCy architecture so program size is overall smaller.

Originally posted by: drag
so your saying that this is just a s/390 series that has a different marketing name?
Click to expand...
Sorta. There were improvements of course. Think of it as kinda like Windows 2000 to Windows NT. We all know 2K was NT 5.1.
 
Im no expert by any means, but my understanding is that the Z architecture is quite the leap from 390?
As in the jump to 64 bit, more/bigger registers, etc?

Kinda like IA-32 to AMD-64?

Not like IA-32 to IA-64, but still, it's a far bigger change than i386 to i486, or s/380 to s/390, no?
 
Originally posted by: Sunner
Im no expert by any means, but my understanding is that the Z architecture is quite the leap from 390?
As in the jump to 64 bit, more/bigger registers, etc?

Kinda like IA-32 to AMD-64?

Not like IA-32 to IA-64, but still, it's a far bigger change than i386 to i486, or s/380 to s/390, no?
Click to expand...

Yea the IA-32 - AMD-64 is probably closer. Its kinda hard to compare to the desktop world because the desktop world is sooooo modular. The thing about the 390/z architecture is its in such a constant state of being 100% backwards compatable, its really hard to say at what point you can really call it a new architecture.

IMHO the easiest way to spot why the name change isn't really for techical reasons is the timing. They changed the name of every one of their other servers on the same day. I can't imagine they all had radical architecture changes as well 🙂

Interesting sidenote, the move was actually from thirty-one bit to 64bit. Earlier frames had to sacrifice that last bit as a way of letting the system know when things were 24bit, and when they were 31. This made for an extremely annoying 2GB of addressable memory. Its also why DASD's are all 2GB. Granted they went to great lenghts to compensate for this with a sort of virtual-memory-within-virtual-memory system, and logical volume management. It got to the point with G6's where the 2GB of main memory was more like a glorified L3 cache.

(oh and FYI, there was no s/380)
 
I'll be damned, I thought they went from 360 to 390, then Z gradually.
Should have read the article I linked to, which does indeed state that they skipped 380.

Oh well...
 
Originally posted by: Sunner
I'll be damned, I thought they went from 360 to 390, then Z gradually.
Click to expand...
They did, sorry when I said "They changed the name of every one of their other servers on the same day." I meant the rs/6000 -> pSeries as/400 -> iSeries thing. A G5 or G6 mainframe is still a s/390.

s/360 was released in the 60's, s/370 in the 70's, s/390 in the 90's, and s/300 in 2000? Uh oh, call the marketing department, we need a new name. Hey "z" is a cool letter you don't see to often.

bart

 
That wasn't the part that confused me, I just assumed they went 360->370->380->390 with some minor changes in between those.

But then, we never got a 80186 or 68010 either 🙂
 
Originally posted by: Sunner
That wasn't the part that confused me, I just assumed they went 360->370->380->390 with some minor changes in between those.
But then, we never got a 80186 or 68010 either 🙂
Click to expand...
Uh, no... 😉
 
Originally posted by: Sunner
I said we never got them 😉
Click to expand...
Well I guess if you want to argue semantics, then we're both right. 😱 😉 😀

<-- had an IBM PS/1 (10MHz 80286 machine) when he was younger; it made a great QuickBASIC workstation. 😛
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top