BC Card, the largest credit card processor in South Korea, is dumping its Unix servers, made by Hewlett-Packard and Sun Microsystems, and moving its processing applications to mainframes made by IBM. No, time is not moving backwards on the Korean Peninsula. But BC Card, which is a coalition of 11 South Korean banks that handles …
Mainframes for Dummies ?
"it may be hampered by a lack of techies with deep mainframe experience."
I always wondered how you manage to get into the mainframe arena. After all, you can hardly pick-up a second hand IBM mainframe on E-Bay for £50. And I haven't seen any suitably titled books in the shops or on Amazon, Compman, etc.
Computers are like cars. You can turbo charge all you like, but you can't beat cubic inches at the end of the day.
Anyone still predicting the death of the mainframe?
There's an embedded vendor (www.mc.com, no relation of mine) who have an informal metric of GFLOPS / cubic foot (maybe TFLOPS / cubic foot these days...) Quite useful to know when you're short on space.
"You can turbo charge all you like, but you can't beat cubic inches at the end of the day."
Until you want to go round a corner. Then all that weight screws your handling.
Although it's mostly Yanks who go for the cubes and they wouldn't know what a decent suspension set-up was if one leapt up and bit them.
Is an embedded vendor something like a built-in salesman?
Did you require surgery to get one?
Sounds nasty to me.
Empty space ain't heavy, it's the stuff around it that is
Not so - the McLaren F1 has plenty of cubic inches (6.1L - can't be bothered to do the conversion, but it's a lot) and no turbo, but apprently goes round corners pretty well indeed. Won Le Mans quite convincingly without wallowing in the corners like an undersprung soggy Wheatabix.
For some good car oggling, take a look at www.jaylenosgarage.com.
The porting aspects intrigue me. I would assume that unless the mainframe is running Linux or AIX, they have a huge lot of code rewriting to do. Or is the code all pure Java, and can mainframes can run Java so well these days that the apps notice no difference?
There's porting, and there's porting.
Most of the apps are existing IBM apps that are available for both MF and distributed systems. Migrating your DB2 databases into DB2 on the mainframe, or deployed your J2EE apps to WebSphere on MF, should be pretty painless.
Just another upgrade
They need more umpf and get some big IBM boxes. But IBM-mainframes run linux these days, so where's the revolutionary change?
I think they've been fooled into the hype that Mainframes are all powerful because they're so expensive, but it's not true and they'll simply regret that choice.
The processor development has not kept pace with the PC, the RAM has, but only because it comes from PC technology, the io tech also comes from PCs. So they're buying mainly PC tech in a large box with a high price tag and piss poor processor units, and a lot of vague buzzwords like 'dynamic', and 'multichannel'.
There is a damn good reason IBM is coy about benchmarking mainframes.
I don't think they've been fooled, and I don't think this is about performance. I think it's about how easy things are to manage. It's hard to get away from the fact that a collection of "midrange" (typically Unix) systems are harder to manage than a much smaller number of Mainframe systems, both because there are more of them, and because the tools are significantly less mature. And I'm saying this as someone who manages Unix systems for his living.
significantly less mature? And how long has Unix been around?
"The processor development has not kept pace with the PC"
Since when did a 5GHz Power6+ not equal the latest and greatest from Intel (3GHz these days?) and AMD. A lots of these banking applications need serious batchjob grunt, and there's some jolly clever bits of architecture in the Power6 et al that were designed specifically for banking applications (e.g. hardware acceleration for BCD arithmatic). That's why they've gone for IBM.
Meanwhile, in the real world...
First, you should probably look up the z6 chip* before you spout uninformed drivel about mainframe processors. Mind you, it's pretty typical of PC nerds to focus entirely on the processor* (even if it is years ahead of x86, which it is, and even if you aren't doing much of the aforementioned focusing because you might know better if it you were) and to ignore the most important part of a mainframe: sustained I/O throughput. At the end of the day, that's what enables you (and six thousand other people in the same second) to swipe a credit card to buy a graphics card for your neon-festooned core i7 Windows gaming rig.
If these things are using "PC technology", then why are things like SMP, virtualization, and crossbar buses still relatively new to PCs? Mainframes did that kind of stuff at about the time when ELO was selling music on 33 RPM vinyl. Where's Intel's quad-core 4.7GHz chip? Come back later when you can point me to a PC that can handle thousands of transactions per second for a decade straight without slowing down or having to be shut off. When people buy mainframes, they pay for I/O and they pay for reliability. "PC technology" is not up to that task, and might get there before everybody in this thread has gray hair.
*A little clarification, especially for all the people I keep encountering who think z/Series uses POWER6 for much other than I/O controllers:
(That was in 2007. See that "4+ GHz" spot there? Already happened.)
Re: Since when?
""The processor development has not kept pace with the PC"
Since when did a 5GHz Power6+ not equal the latest and greatest from Intel (3GHz these days?) and AMD. A lots of these banking applications need serious batchjob grunt, and there's some jolly clever bits of architecture in the Power6 et al that were designed specifically for banking applications (e.g. hardware acceleration for BCD arithmatic). That's why they've gone for IBM."
Mainframes do not use Power processors. They have their own distinct processors. Here is a presentation about what is in a z10:
"So they're buying mainly PC tech in a large box with a high price tag and piss poor processor units, and a lot of vague buzzwords like 'dynamic', and 'multichannel'."
You don't buy a mainframe for its FLOPS performance , you buy it for its I/O performance , scalabilty and reliability. Main frames are used in arenas where you have ZERO downtime. The box runs 24/7/365 and any maintenance is done on the fly. Try that with a unix or windows box.
All unsupportable claims. There's no special magic in Mainframes, it's just generic technology and out of date processors packaged up as something special by the few remaining vendors that ship systems marked as 'mainframes'.
They don't have any special technology there, and Mainframes don't run applications any better than any other simple OS.
All buzz words and salesmanship, and avoiding critical independent benchmarking against comparable PC servers.
Try that on a unix or Windows box....?
Every day, old chap. Virtualise the Windows and/or Unix installations onto a clustered affair (VMware for example), and away you go.
So, a bit like mainframe, only a bogload cheaper (..... so mainframe for the rest of us).
Google Pixar etc.
The brutal reality of this is:
Software on Mainframes crashes at least as often as BSD if not more, fewer users mean fewer eyeballs have used the code and it is likely to have more bugs as a result. i.e. Websphere is likely less bug free than Apache for a similar maturity due to fewer users and fewer apps that use it.
Forget the linux slices on mainframes, they are crap, your software will crash as often, plus every now and again the mainframe will kill the whole linux slice thinking there is a problem.
Mainframes cannot offer stability so they offer 'availability', i.e. when an app takes too much processing, or disk space are whatever, it is killed and a fresh instance started. Turning a *maybe*-crash into a *real* crash.... ahh but there is another instance running in a different slice, so it's 'available' still..... well at least a new instance of it is...
IBM can't compete in the processor market, the processors in a mainframe are not better in any sense than their PC counterparts. Likewise the rest of the kit, mostly stock technology from the PC world.
IO? Again IBM do not dominate the IO world, Pixar will not switch its render farm to a mainframe, even the storage will not be switched.
Scalability? Google will not switch their search engine to a Mainframe.
Dependability?... NYSE won't switch back to mainframes.
Maturity, yes, I'll give you that, I am not an early adopter of anything, no-one should depend on something till lots and lots of user eyes have given it a clean bill of health. Likewise really old software suffers because the hardware has changed, he protocols changed, but the eyeballs don't take a second look at the new combination of new hardware plus old software. So well tested software on the older mainframe hardware it was designed, is similar to well tested software on older PC hardware it was designed for.
So if there are 500 customers of a platform, vs 5 million, then 10000 times the eyeballs. The world is not favorable to mainframes.
So what do we have? A platform with few users, few developers, no special technology, few real world reviews, but an awful lot of vague sales twaddle. I don't know what the back story is here, but this bank has made a mistake in my view.
Funny you should mention it.
Virtualize. You mean like mainframes have been doing since the Nixon administration? VMware is a replacement for z/VM like Segway Inc. is a replacement for Kawasaki Heavy Industries.
There's so much shit in this post that you could smell it on Kepler 7b. I'll happily accept that Websphere and Tivoli(*) products stink, but the rest of these assertions are just utter crap. Come back and post some real world information after you've been doing this sort of thing for 10+ years (Linux on the mainframe) or 27+ years (virtualization).
(*) Obligatory Tivoli bashing as they refuse to accept that zVM exists - morons.
Aside: why are all the people who are bashing this decision in this thread, anonymous cowards?
after reading TFA, it seems that they won on the idea that they only pay for what they use.
going through a slow period, shut down bits of the mainframe, reduce licensing cost.
going through a period of high use? pay to unlock more processor to git-er-dun.
has to be one of the largest pay-as-you-go tariffs i can think of!
i guess the previous setup was like a fixed rate contract - had to have enuff free minutes to cover your useage in a worst case scenario, but if they weren't used, then they were paying for something they weren't needing.
it's all economics really. something you'd hope banks would be good at.
> and Mainframes don't run applications any better than any other simple OS.
Try this on Linux:
-- start with Linux and Apache, database server (etc) and a set of running applications on a live machine
-- It's been like that for months, but now it is time for some software upgrades. So:
-- now start a hypervisor under Linux
-- now swap the hypervisor and Linux so Linux is running under the hypervisor
-- then start a second Linux (perhaps a newer version) under the hypervisor
-- run Apache, applications etc under the second Linux, again perhaps newer versions. Run as many acceptance tests as you like
-- when you are happy, instruct the old applications (etc) to hot-switch in-flight transactions to the new Linux version
-- wait a minute or two for everything to hotswitch over
-- Leave the old stuff in place (but in sleepy mode) for a few days, until you are happy the new stuff does not need a hot fallback any more
-- stop the old applications, Apache, etc
-- shut down the old Linux
-- swap the hypervisor and new Linux so the Linux is running the machine
-- shut down the hypervisor
You have now upgraded your operating system, middleware, applications, etc. They started running on bare metal for max processing. And they end that way too. No application was ever down from a user perspective; no one had to sit through a reboot, or log out and back in again
The only possible noticeable change is your response times may have been slightly slower during the parts of the process (but you probably turned a spare CPU on to cover that, so perhaps not).
When you something as basic as that on a non-mainframe platform, we can start discussing the things that Mainframes do really well.
"I always wondered how you manage to get into the mainframe arena. After all, you can hardly pick-up a second hand IBM mainframe on E-Bay for £50. "
Simple. Shaums Outline Series "IBM Programming."
Not (as many an unwary punter has thought) x86 machine code but the actual 360/370 assembler.
As an aside during the mid 80's IBM did a funny AT which had a special board in it running a S370 processor. Written up in Byte (cannot remember when) and never a bit hit.
There are specialist mainframe training courses but they are not particularly cheap.
With high level clearance you can write a very high efficiency scheduler by creating the mother of *all* TSR programs.
You can guess what's in my pocket.
It happens by accident, got a job in a bank as an IT grunt and started on mainframes that way. One day normal server techy next you find this whole new world of green screens and no GUI's.... its a strange old place.
Right tool for the right job
Suprising that people think mainframes are more expensive, windows fanboys never account for the hidden cost in their systems, downtime to the business, support costs etc. Worked on mainframes for years and can only think of one occasion where there was any down time, and that was because we were running an unsupported version of IMS. Never had to worry about disks breaking, cpu's frying etc that was all done on the fly and invisibly to the end user. If there is a problem with the mainframe it will call the IBM support engineers and he will be on site to replace the part in real time.
Despite what any of the windows fanboys might say, mainframes are used because they are reliable. They dont have downtime, run 24x7 and have failover systems that you simply dont get in windows systems. Yes google runs on a vast array of utility pc boxes but if google stops working for a short while you wont be stuck at the tesco checkout with your weekly shopping wondering why your card payment has been declined.
Use the right tool for the job.
Horses For Courses
About time that someone used some common sense. Mainframes have come down in price because they had to, but they are still far more reliable than any server farm. I come from a mainframe background, mostly airline and we used to have over 99.9% up time over a year. Try and match that you Windows fans. Whoever mentioned green screens is definitely living in the past. These days, mainframes will support any sort of GUI front end that you like even Windows if you have been deluded enough although Linus would probably be a better and more reliable bet.
It used to make me laugh when you got a bug in your Windows application and support told us to re-boot. When we got an error on the mainframe software which we had written, we told and not bery politely to go fix it, which we did.
I'm not saying it was perfect, far from it, but the underlying operating system and hardware was ultra reliable.
Servers are useful for running many things but the bottom line is that they are not the ultimate solution.
Managers who make these decisions (and governments) should learn that computers are merely tools to aid what you are trying to do not the end in and of itself. In other words technology for technology's sake is NOT the correct decision, just as servers are not necessarily the answer just because they are the latest fashion.
I repeat - Horses for courses!!
Re: Right tool for the right job
Funny that you would state how great Mainframes are compared to PC's. The article is about replacing Unix boxes, not mainframes.
Mainframes are slow. That's why IBM will not release any industry standard benchmarks for them.
Of course it seems from the article that the mainframe is fast enough. The problem for Unix vendors is that mainframes are hella reliable. Of course they are not very easy to use and require over paid "Systems Programmers" to use and a plethora of operations people to manage. To run a proper Unix box, the same can also be said, but there are more Unix people out there so they are a bit less overpaid...
Oh well, I'm blathering now.
It's about scale
We just consolidated 400 Sun servers onto two maxed p595s and dropped our IT cost by $5M/yr. In another five years, we'll probably have no choice but to move to a mainframe. Either that or start ballooning again and create a farm of 595s. Mainframes scale better and are much more cost effective. Some workloads will never run well on even the most powerful UNIX boxes. The kiddies, who live their lives on x86 gear cannot understand that the mainframe will never go away.
Yes, the CPUs are dog slow. But they are reliable. That is why you buy them. Not because of performance.
You can emulate one Mainframe in software on a x86 PC with "Herkules", in fact, the emulator is now sold to companies that need a Mainframe backup, in case their Mainframe crashes. You dont easily buy another Mainframe as a backup, they are too expensive. According to Wikipedia, Tom Lehmann, cofounder of TurboHerkules explains that you can emulate a Mainframe worth of 3.200 MIPS on a 8-socket Nehalem box. That is 400MIPS per Nehalem CPU.
Current highend z10 Mainframes offer you 28.000MIPS (using 64 of those dog slow CPUs). That is 437 MIPS per z10 CPU, slightly faster than one Nehalem CPU:
But the Nehalem is giving you 400MIPS of power under software emulation! Software emulation is a factor 5-10 times slower. If one Nehalem could run the IBM Mainframe programs in native code, it would give 2.000-4.000MIPS! It would suffice with 7-14 Nehalem CPUs to match one fully configured highend z10. Two 8-socket Nehalem PCs have much higher performance than one z10 with 64 CPUs.
How much does one fully configured z10 with 64 CPUs cost? $50 million? How much does two 8-socket Nehalems cost? $25.000 each? You do the math.
Yes, the Mainframes may be reliable, but they are dog slow. Any modern x86 CPU (such as Nehalem) is 5-10 times faster than one state-of-the-art z10 CPU.
- Geek's Guide to Britain INSIDE GCHQ: Welcome to Cheltenham's cottage industry
- 'Catastrophic failure' of 3D-printed gun in Oz Police test
- Game Theory Is the next-gen console war already One?
- Apple cored: Samsung sells 10 million Galaxy S4 in a month
- BBC suspends CTO after it wastes £100m on doomed IT system