back to article IBM heaves new System z minis at mainframe shops

In the hope of continuing the System z upgrade boom that started last summer, Big Blue has rounded out its lineup with a midrange – what IBM calls "Business Class" – mainframe, the System zEnterprise 114. The machine is more or less what the rumors of two weeks ago hinted it would be – although it turns out there are two models …

COMMENTS

This topic is closed for new posts.
Silver badge
WTF?

What?

"ranging from a low of 26 MIPS "

Am I missing something here, but as far as I can see my HTC smart phone has more than 26MIPS of processing performance free to run any apps, in fact, probably almost 10 times that.

Just who would spunk several tens of thousands of dollars on a sub-early-90s-486 performance machine, even if it is highly redundant?

0
4
Thumb Down

Re: What?

"MIPS" in the context of mainframes is a specific benchmark, only directly comparable to other hardware of the same type. z MIPS do not correspond to anything like "millions of instructions per second" on x86, ARM, or anything else of the sort. z MIPS don't even correspond to MIPS on other mainframe platforms, like OS 2200, MCP,, and others.

0
0

Count not MIPS

Simple explanation. An online system like your phone spends most of its 'instructions' doing pointless, redundant non-work actions, because its workload is unpredictable and most of its existence is spent polling and caching, while awaiting real work. It has to be like this, in order to be responsive - but your battery life will attest to how piss-poor it is, as an actual 'computer'. Try running your payroll through it, and there'd be no money left to pay anyone, once the electricity bill was accounted for.

A mainframe, on the other hand, performs certain high-load tasks - such as moving pages from memory to physical media - using some 5-10,000 less 'instructions' than your phone will. It can do this because it is designed to - because it doesn't need to be responsive: just fast. The number of clock cycles, per instruction is far higher, you see, so the workload is blisteringly fast, but the number of instructions needed, is low. Who needs instructions, when the job is so predictable?

So count-not MIPS. Measure work. In truth, yes, are missing something; but it's only mainframes, so you're not actually missing much. There's no Angry Birds on a mainframe.

1
0
Silver badge
FAIL

@Count not MIPS

Imagine how foolish I now feel thinking that MIPS stood for "millions of instructions per second" (or integer operations), like every other computer text book uses. Doh!

"An online system like your phone spends most of its 'instructions' doing pointless, redundant non-work actions, because its workload is unpredictable and most of its existence is spent polling and caching, while awaiting real work."

Er, no, any sensible CPU that aims for low power use (and that includes some DSP) have an idle instruction that powers down until an interrupt awakens it (event or timer). So they don't do pointless work, they do real work while needed, and power-save the rest of the time.

"such as moving pages from memory to physical media - using some 5-10,000 less 'instructions' than your phone will."

Er, that also sounds like bollocks, as even the pitiful x86 only takes a dozen or so clock cycles to do a memory copy, less per word for repeated 'string' operations, and for large block transfers you have DMA. Even on a basic computer.

So can we have a *meaningful* explanation of what a z-series MIP is, and how it compares with actual execution throughput?

0
1

This post has been deleted by a moderator

Silver badge

@Daniel 1

Seems your cage was rattled today - resorting to ad hominem arguments?

0
0
Coat

@NO

"So can we have a *meaningful* explanation of what a z-series MIP is, and how it compares with actual execution throughput?"

Answer is no because IBM don't want people comparing Z to Power/Linux/Windows - I'm sure IBM's sales team would respond as rudely as Daniel 1 did :-)

To be fair you cannot compare - I've done some performance estimation on Z and for unix'y like things which run under unix system services - emulation like layer - its not very good so for running Java I'd not use it unless the zAAPs are really cheap. For IO and DB like workloads or native apps then its blisteringly fast... So impossible to compare in a MIPS/dhyrstones/linpack like measure of CPU performance.

In fact the MIPS rating of the machine changes depending on the workload....

0
0
Joke

Ad-hominem?

More of a Oh Gofuckoffinem argument...

0
0
Anonymous Coward

Why on earth should he?

The comments section of this story consists of you saying your phone is as good as a mainframe and demanding people explain why it isn't. If you're really interested in mainframes, read the book. If reading a book sounds like a bit too much work, play with your phone. It's nobody's job to enlighten you. Welcome to the 21st century, please form an orderly queue leave your sense of entitlement at the door.

1
0
Silver badge
Happy

RE: Why on earth should he?

He could even go download the Kindle app for Android and then go read the book on his phone.

IBM are notoriously vague when it comes to trying to quantify the performance of their mainframes as they really don't like comparing even to their own Power kit. However, seeing as most of their mainframe customers are locked in because of historic application choices rather than becuase they think mainframes are better than anything else, IBM are able to get away with it. Of course, when other companies come up with solutions that allow mainframe customers to try mainframe apps on non-mainframe kit (think Platform Solutions Inc.) and realise how much IBM were rooking them, IBM fires up the lawyers and/or buys them out (http://www.theregister.co.uk/2008/07/02/ibm_buys_psi/).

0
1

OK, basic way of explaining

Home high end PC=Toyota sports model

Home PC=Toyota ordinary model

Business server (x86)=Pickup truck

Mainframe=18 wheeler+another one for redundancy, they even go same place, in parallel to verify the load has arrived.

0
0

readme.txt

read this first: SG24-6366-02

then understand.

then post.

1
0
Silver badge

@readme.txt

792 pages - hardly a "readme.txt" - and for a first glance no meaningful comparisons with any other systems, though 0.91 watts per MIPS was mentioned (which seems not that spacial compared to ARM-based hardware).

Maybe you could enlighten us?

0
0

see other posts

SG24 was just mentioned in order to give people a meaningful start about z.

0
0

Simon Say... Do Nothing

In simple terms, 'Simon Says do nothing' is an 'instruction'. Mainframes are all about removing the 'Simon Says do nothing' instructions, and leaving in the 'Simon Says... Read the Complete Works of Shakespeare' instructions.

Most of your phone's battery life is burned away doing nothing.

Because it was told to.

Because that make sense, in a phone.

0
0
Silver badge

IBM MIPS = Meaningless Indicator of Processor Speed

It's simply a mechanism for IBM to adjust the amount they charge you for software according to the 'power' of the system on which you're running it. So it's vaguely comparable across the IBM system range, but nothing whatever to do with MIPS on a PC.

0
0
Silver badge

@Chris Miller

Thank you for an understandable translation.

0
0

IBM's mistake is

They shouldn't use MIPS term to non mainframe public who always misunderstand.

Can't they do a good PR video for general tech public even including lots of developers answering the basic question? "why do banks, governments and large enterprise require a multi million mainframe?"

I guess a lot of people think "because they are stupid." and it isn't the case.

Isn't the "transactions per second" a better benchmark for a mainframe?

0
0
Holmes

Actually it does mean MIPS

It's just it's unreliable to actually compare to anything else. First because ALL mainframes run a form of hypervisor, some OS's actually run under two hypervisors. Second, almost all I/O is not done by the main processors; in fact there are a great number of processors in the system that handle I/O once it's made its way through the GPU. Third, IBM System Z uses a form of macrocode or VLIW in its firmware which allows direct calls from an app or OS to bypass what would normally be result in a series of instructions that might cause interrupts in the GPU(s) and calls back to the OS in other environments. So apples and oranges.

So, MIPS are indeed measure of millions of instructions per second, they are just not comparable, hence they are meaningless. However, if you want to compare a specific app running a real world workload using say 4x cores, 128Gb of memory, in a VM doing say 250 IOPS per second, with a failover server and networking, you can come up with some pretty compelling price alternatives to mainframes. Are they valid, are they comparable? It depends on the business need?

Beauty as they say, is in the eye of the beholder, price and value are the same. I work at Dell (now)

1
0

@Paul Crawford

You can not just use IBM Mainframe MIPS to compare against normal x86 MIPS. So that is not really comparable. Also, IBM marketing will try to make it difficult to compare performance between x86 and Mainframes. For instance, it would be easy to run Dhrystones / SPECint / SPECfloat / any other benchmark on a Mainframe, but you will never see such benchmarks published. IBM marketing would never allow publication. But we know that whenever IBM has anything advantageous to say, IBM says so. So, why are IBM not publishing common benchmarks run on IBM Mainframes? IBM Mainframes crush everything in terms of performance, right?

.

The new IBM Mainframe Z196 cpu, is the "Worlds fastest cpu" according to IBM:

http://www-03.ibm.com/press/us/en/pressrelease/32414.wss

It has crazy specs. 5.26GHz, ~300MB cache (L1+L2+L3). Close to half a GB of cache! So, how fast is it? Let us see. It is 50% faster than the previous Z10 cpu. How fast is a Z10 Mainframe cpu? The biggest Z10 Mainframe with 64 cpus, give you 28.000MIPS. Now, that is a lot! How much is it?

.

If you use an 8-socket Intel Nehalem-EX server, and use the IBM Mainframe software emulator "TurboHercules", then you get 3.200MIPS:

http://en.wikipedia.org/wiki/Hercules_%28emulator%29#Performance

That is under software emulation, which is 5-10x slower than running native code. So, an 8-socket Nehalem-EX server would actually give 5-10x higher performance if you ported the Mainframe applications, so we could run native code. That is: 16.000-32.000 MIPS. Thus, an old 8-socket x86 server would give you as many MIPS as the biggest Z10 IBM Mainframe with 64 cpus. If you build an Emulator for IBM Mainframes, then you know the ins and outs of Mainframes.

.

Another developer that ported Linux to IBM Mainframes came up with this rule:

1 IBM Mainframe MIPS == 4 MHz x86.

http://www.mail-archive.com/linux-390@vm.marist.edu/msg18587.html

So, the biggest IBM Mainframe Z10 with 64 cpus, correspond to 28.000MIPS == 112.000MHz x86. But, a single 8-core Intel Nehalem-EX running at 2.5GHz, has 20.000MHz in total. Thus, you would just need a few Nehalem-EX to reach 112GHz. Again, we see that you need a few Nehalem-EX to beat 64cpus of the Z10 Mainframe.

.

And, Here is a source from Microsoft

http://www.microsoft.com/presspass/features/2003/sep03/09-15LinuxStudies.mspx?

"we found that each [z9] mainframe CPU performed 14 percent less work than one [single core] 900 MHz Intel Xeon processor running Windows Server 2003."

The z10 is 50% faster than z9, and the z196 is 50% faster than z10, which means a z196 is 1.5 x 1.5 = 2.25 times faster than a z9. This means a z196 corresponds to 2.25 x 900MHz = 2 GHz Intel single core Xeon. But todays modern server x86 cpus have 8 cores, which means they have in total 8 cores x 2 GHz = 16 GHz. Again we see that you need only a few 8-core Nehalem-EX to match the biggest z10 Mainframe.

.

Now the Z196 cpu is 50% faster than the z10 cpu. Instead of using 8-socket x86 servers, you need to use 12-socket x86 servers to beat the biggest Z196 IBM Mainframe, utilizing the "worlds fastest cpu". In short, if you can migrate your workload to x86, then do it.

.

.

Sure, IBM claims their Mainframe can virtualize 1.500 x86 servers. If you read the entire study, you will see that IBM assumes all x86 servers to idle, and the Mainframe is 100% loaded! Now what happens when a powerful x86 server does some work? The weak Mainframe can not virtualize that load. The Mainframe cpus are to weak.

.

Sure, IBM claims the Mainframes have brutal I/O and can handle 400.000 logged in users. But I would not be surprised if 95% of the users idle and only a few of the users do some work.

1
3
Silver badge

@Kebabbert

Thanks for providing much of the missing information on the 'mainframe MIPS' question.

I am somewhat surprised by the almost religious reaction of some folk to pointing out that the option to buy 28MIPS, in the classical sense, is a piddly small amount of CPU performance. Yes, in 1989 it might have virtulised a server, but today it is nothing (even accepting the joys of crypto accelerators).

So from the helpful comments here it seems there are limited reasons for wanting to put one's financial balls in an IBM vice. If the performance was so good in terms of £/{standard load measure} it would be well advertised, and the lack of the likes of SPECint/linpack/etc results tells me a lot (at least for my sort of CPU use).

0
0
FAIL

@Kebbabert brain fart

Do you not engage your brain before typing out a load of daft anti-IBM propaganda? If even a tiny amount of what is in your post was true then do you not think everyone would move their workloads to x86 servers? Especially given the astronomical costs of the mainframe environment?

Maybe there are a few things you are not considering....

As an aside in the real world (Not a place for a Kebbabert to tread) a large banking group I know have a large chunk of their real estate on x86, IBM Power and IBM Z -- of the three it is quite clear which gives them the best bang for the buck as both the x86 and Power estates are thousands of machines which are an utter nightmare to maintain. Z runs a very large proportion of their critical apps in a single managed environment that is easily distributed using GDPS. I think Kebby would get feck all commission trying to wean them off the 'weak' mainframe CPUs....

0
0

@Oolons

"...Do you not engage your brain before typing out a load of daft anti-IBM propaganda? If even a tiny amount of what is in your post was true then do you not think everyone would move their workloads to x86 servers?..."

Again, let me repeat, I am only saying that Mainframe cpus are very weak. The evidence are numerous, as I have provided several links from Mainframe experts. If you have written an Mainframe emulator, then you must be considered an "expert", yes?

Because I have provided lot of links and proofs, my post can not be considered as "propaganda". In fact, my links are true. I am not trying to spread false rumours, making unsubstatianted claims. What I say is true, there is major substance in my claims. The IBM Mainframe cpus are very weak. Just look at all the links I provided.

If you have counter proofs, then I will gladly read them. If you can show benchmarks that show how fast Mainframes are, then I will of course stop say that Mainframe cpus are weak. But, there are no such benchmarks. Why? Maybe there is a reason that IBM never publishes Mainframe benchmarks? Have you thought about that? Maybe the links I collected here, are correct? Maybe you are fooled?

.

.

"...As an aside in the real world (Not a place for a Kebbabert to tread) a large banking group..."

I don't work in the banking world. That world is boring. Come on, Mainframes?! How sexy is COBOL? Who wants to work with antique and slow hardware such as Mainframes? What does a mainframe do? It updates lot of accounts with a % batchwise. Do a trivial calculation on lot of bank accounts. How sexy is that?

I work in finance. Extremely low latency, very high throughput, extreme cpu demands, algorithmic trading, High frequency trading, risk calculations, etc. Now, that is sexy! Extreme tech. Lastly I heard about HFT traders using FPGA to implement their algorithms! Apparently FPGAs can lower the latency 10x or so. That is cool and exciting indeed. Forefront tech. A Mainframe is not used for such things, and could never do that. Too slow, and bad latency.

There is a reason stock exchange systems are not run on Mainframes. If a exchange system crashes, the exchange will loose lot of money, and customers (hedge funds, IB, etc) will start to trade on other exchanges (they can not sell/buy when they see an opportunity and loose lot of money, often they leverage 5x or more). If Mainframes were fast, and reliable (I have several links that show Mainframes to crash, one reason the software emulator TurboHercules is sold is because customers need failover when their Mainframe crashes) then the finance world would not care how much a Mainframe costed. The Wall Street firms would get Mainframes no matter the price. But guess what? Wall street firms are not using Mainframes, they use Unix and Linux. Why is that? Maybe you should start ask you some questions?

Only the slow banking world, without the need for extreme performance use IBM Mainframes. Finance does not.

0
1

@Paul Crawford

"...I am somewhat surprised by the almost religious reaction of some folk to pointing out that the option to buy 28MIPS , in the classical sense, is a piddly small amount of CPU performance"

Yes, but dont say that to them, or you will upset them. Look at my post. I provided several links to Mainframe experts, and the IBM supporters gets crazy and call me names "Kebab fart/Kebab brain/etc" and say that I spread false rumours and FUD, "anti-IBM propaganda", etc. Just see below in the comments here. I never make unsubstantiated claims, I always back up my claims with links. I have even got complaints that I post too many links (which proves my claims) - the IBM supporters would prefer me to not post links, so they can call my posts to be false rumours and FUD. But alas, I will continue to post links upsetting them even more.

Be careful with the IBM supporters, they are FUDing a lot and will call you names if you say something disadvantageous to IBM. According to wikipedia, IBM is the first company to employ FUD on a wider basis, and is famous for doing that:

http://en.wikipedia.org/wiki/Fear,_uncertainty_and_doubt#Definition

The IBM supporters are just following the same tradition. So be careful. For instance, there were lot of posts here that they migrated from HP-UX and Solaris to POWER7 servers, but the POWER7 servers were not even released yet.

But I dont care, as long as I speak true and can post to links, I will do that. If I can not post links then it means I am spreading false rumours, which I never do. For instance, when I say that "IBM will kill off AIX", the IBM supporters get crazy:

http://news.cnet.com/2100-1001-982512.html

and say I spread false rumours and FUD, even though I post links that support my claim. That only proves their religious reaction.

1
2
Bronze badge
Mushroom

Excellent comments

I've been comparing System Z with HPC Intel based grids on a cost-per-crunch basis and I've come to similar conclusions (Whetstone MIPS is a benchmark that runs on many platforms) about mainframe CPUs. There's nothing magic about the silicon just because it runs a different OS.

That said, I come from a mid-range (mostly *NIX) background and I've come to love mainframes for their reliability under sustained load and their resource management. That's all down to the OS. The latest AIX platforms are starting to show some of the same tricks but mostly software licensing lets them down and prevents the full versatility being used.

Why do traders run on mid-range? Mostly because they scale out (not up) much more economically and the skills to develop on them are cheaper.

I hope that some of those criticising your posts will also link to articles demonstrating their points. I want to learn more about performance comparisons cross platform as it's a subject that's under reported.

0
0
Silver badge
Happy

RE: @Paul Crawford

I think Kebbie is hoping IBM do a PSI on him and buy him out!

0
1
Holmes

Jup, they will make him.. mainframe capacity planner.

Most likely not...

But when you've been involved in 'downsizing' projects where people have moved from HP3000 or IBM Mainframes or older platforms, you get a certain respect for these old platforms.

I've seen people size based upon the assumption that Java code could do just as many business transactions as highly tuned old Cobol would be able to do.

But then again, IT infrastructure is simple to people who do by looking at product briefs.

// jesper

1
0

@Jesper Frimann

I am focusing on CPU performance in the post above, not I/O. So, when I say that you can download TurboHercules and use it to virtualize Mainframes, I speak about CPU performance. An x86 server would never be able to handle the I/O of a Mainframe. I have never questioned a Mainframe's ability to do superior I/O. I have heard that a big Mainframe has 296.000 I/O channels.

If you focus on Mainframe CPU performance, then you can emulate the work load on an x86 server.

If you focus on Mainframe I/O, then you can not emulate it on an x86 server. Then you need to stay on Mainframes.

.

But traditionally, Mainframes are used in banking: Update large number of accounts with a simple calculation (increase interest rate). Lot of accounts involved. Batch work. Much I/O. Cpu performance is not the focus.

In finance and in stock exchange systems, you typically use much cpu performance. And also I/O of course. But larger systems are distributed so there is no single server that has to handle lots of I/O. For instance, risk calculations take much cpu. And also MCMC methods.

0
1

Specialised

For example, mainframes doesn't spend any CPU for rounding money, a specialized processor "in book" does it. They also natively support decimal numbers which are used in finance.

I am not in mainframe business, but I just can't stand to people who really think all fortune 500 IT managers are either stupid or must be taking bribe to keep buying mainframes. Their large shareholders must be also stupid so they don't sue the company for wasting money on dinosaurs.

The Register is one of few general IT news sites who features these rare monsters and actual unix servers. Hope people writing these stories does ignore comments section.

1
0
Holmes

Re:Kebabbert.

You didn't really read and understand what I wrote now did you ?

My point was that raw processor performance is just one of many factors that governs how many 'business transactions' you can do on a server.

// Jesper

0
0

@Jesper Frimann

"...You didn't really read and understand what I wrote now did you ?..."

And I must ask you the same thing: you didnt really understood what I wrote, did you? I am ONLY claiming that Mainframe cpus are weak, and have provided numerous evidence for that. That is all I claim.

You mock me for believing an x86 server could replace a Mainframe. No, that is not what I am saying. There are I/O intensive workloads where an x86 can not compete with a Mainframe, in that case you can not replace a Mainframe. Why are you trying to make me sound as if I claim a x86 can replace any Mainframe work load? To make me look dumb?

Let me repeat, in case you did not understand, the Mainframe cpus are weak. The evidence are numerous. If your Mainframe workload is mainly cpu intensive, then maybe you can replace it with x86 running an software emulator. That is all I say. Please stop making me say things that I have never said.

0
1

@IIgaz

"...For example, mainframes doesn't spend any CPU for rounding money, a specialized processor "in book" does it. They also natively support decimal numbers which are used in finance...."

In that case, Mainframes are doing it wrong.

I work in Finance, and we never round off. We always use whole numbers and book keep the number of decimals separately. We never round off. If Mainframes are rounding off, it is another reason to not use them in finance.

Finance have high requirements of performance, clearly Mainframes can not reach them. They have too high latency, and they have weak cpu performance. Wall street firms pays whatever price if the hardware helps them earn another billion, but no wall street firm is using Mainframes.

Mainframes are used back office: administer all the accounts. Bank world.

Unix/Linux are used front office: trading, earning the money. Finance world.

Finance is sexy. Banking is boring (antique servers and antique COBOL).

0
1
Facepalm

*boggle*

"I work in Finance, and we never round off. We always use whole numbers and book keep the number of decimals separately. We never round off. If Mainframes are rounding off, it is another reason to not use them in finance."

You lack of knowledge in the field where you claim expertise is stunning.

// Jesper who actually have written finance applications many years ago.

1
0
This topic is closed for new posts.

Forums