IBM Leaks Details on New Mainframe

Mark writes "Big Blue inadvertently revealed details about its new z10 Enterprise Class mainframe set to launch on Feb. 26, as well as details on z/OS v1.10, a new version of the mainframe OS due out in September. 'According to an internal IBM document obtained by SearchDataCenter.com, the z10 Enterprise Class will come in five different models and feature 64-way chips, compared with the 54-way z9 mainframes and earlier 32-way models. In a conference call last month, IBM CFO Mark Loughridge told investors that the z10 would have 50% more capacity, which indicates that it will probably tap out at around 27,000 million instructions per second (MIPS) at the top end, compared with about 18,000 MIPS on the previous z9 Enterprise Class.'"

Imagine...

[FatSean]

...how many Linux VMs could be run on one of those things!

This might be a dumb question...

[TubeSteak]

How come they talk about thousands of MIPS instead of just saying GIPS?

Low-End Port to PowerPC?

[Doc Ruby]

These mainframes use the z6 CPU [wikipedia.org], which is closely related to the POWER6, which is closely related to the PowerPC.

Is it at all possible to automatically port any nontrivial z6 software to PPC, if it doesn't require the actually different HW of the z6 (or its much higher performance)? Any possibility to run PPC SW on a z6, with some automatic porting for the higher performance?

Re:Kinda slow, eh?

[ModestMotorhead]

Modern mainframes can run not just z/OS but many other operating systems (like Linux or AIX for example). One of the benefits to mainframe architecture is that it is extremely easy to scale vertically. So, if you want to add more processor to an instance, but add more MIPS. I know, it isn't THAT easy, but depending on the licensing that oyu have from IBM it equates to much easier expansion.

Re:Nah Dried off?

[davidsyes]

I'll wait til it's dried off. Can't have a server that someone took a LEAK on...

Oh, maybe they made early RELEASE of details... I wonder, in IT context, how a vendor can "leak" its own details...

Re:This might be a dumb question...

[Anonymous Coward]

because MIPS means Millions of Instructions Per Second, not Mega Instructions Per Second, so it'd be BIPS if anything.

Also, thousands of millions is less ambiguous for those who don't realise that the UK stopped defining a billion as a million million fifty years ago.

Re:Imagine...

[quanticle]

Splitting one of these mainframes into multiple Linux VMs is actually one of the more common uses for these things.

Re:Imagine...

[Anonymous Coward]

How about a system i worked on where we run 200 Linux system with Oracle DB (prod, QA, Dev...) on three mainframe CPU. In that config. you pay for only three CPU worth of Oracle licenses and you can buy a lot of hardware on the software and support savings....

Actually, in lab tests, the mainframe virtualisation engine went up to 96,000 Linux images on a fairly old mainframe version .... Obviously, this was not a performance test...

Re:Kinda slow, eh?

[lgw]

Google style clustering, where you know some of your hardware will fail from time to time and you're just OK with that, is the first promising alternative to mainframe uptimes since the days of VMS clusters. The best hardware from Sun or HP never came close.

The thing is, it's really hard to write code for a "soft cluster". Being fault-tolerant in your software instead of your hardware is decidedly non-trivial. With a mainframe you just write a check with enough 0s. That's very appealing unless, like Google, you're developing everyhting from scratch anyway.

Re:Kinda slow, eh?

[rbanffy]

Shouldn't the branch predictor be different since the instruction set is different? Or it does its magic on the micro-instruction level (if there is one in P6 and z6)?

BTW, the first message was kind of a joke. It is possible to make more or less the same chip look like a few different ones by changing just a few small parts.

IIRC, there is a company being sued by IBM for making custom-microcode-Itanic-based servers that look too much like IBM mainframes.

Re:Kinda slow, eh?

[Anonymous Coward]

Yeah that's why even Deutsche Bahn was able to replace 300 servers with one mainframe. Far too many people learn about the design and power of mainframes by reading pc magazine. Even if there was a performance hit the stability is worth it to any operation that requires IT services to bring in revenue.

http://www.theregister.co.uk/2005/03/31/deutsche_bahn_ibm_suse_server_consolidation/ [theregister.co.uk]

Re:Kinda slow, eh?

[jimicus]

Google style clustering, where you know some of your hardware will fail from time to time and you're just OK with that, is the first promising alternative to mainframe uptimes since the days of VMS clusters. The best hardware from Sun or HP never came close.

Google have a very specific set of applications which, for the most part, don't really care if chunks of data from the database go missing occasionally, can be easily mirrored and it's not particularly crucial that every mirror is in perfect lockstep.

Try proposing a system like that to the IT (or, for that matter, the Finance) director of a $multi-million firm and let me know how you get on.