IBM Leaks Details on New Mainframe 185
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.'"
Re:54 way chips? (Score:3, Informative)
Re:54 way chips? (Score:5, Informative)
Re:This might be a dumb question... (Score:4, Informative)
Re:Low-End Port to PowerPC? (Score:5, Informative)
The things they share are not visible to the user as they are hidden behind the instruction decoder. You can see some evidence of the fact that IBM are trying to lower costs by sharing a lot of the design between the two lines though from certain new additions to the POWER instruction set, such as hardware support for Binary Coded Decimals (useful in high-throughput financial systems and present in the mainframe line since the 1401 and 700-series, which preceded System/360).
Re:Kinda slow, eh? (Score:4, Informative)
Re:Kinda slow, eh? (Score:5, Informative)
They also support partitioning on the hardware level, so you can run z/OS or Linux virtual machines with almost no overhead (something you've been able to do since it was called System/370). You also have a huge amount more fault tolerance with a system like this (take a look at how many transistors on the CPUs are dedicated to error checking, and then start looking at the peripheral systems).
Re:This might be a dumb question... (Score:3, Informative)
As an added benefit, you avoid the question "which one, metric or imperial?" (no, I'm not going to try to determine the airspeed of an unladen metric tonne), which might come up as the spelling difference isn't significant enough in pronunciation.
Re:Kinda slow, eh? (Score:5, Informative)
Your math is also way off if you think 4 x86 cores outperform this. I'll leave you to do the proper calculations as your homework.
Re:Kinda slow, eh? (Score:5, Informative)
We use them due to ability operate in something called a sysplex. A sysplex is when multiple mainframes share data (known as DASD) and work together. When a mainframe is in a sysplex, you can do all sorts of things to the machine without having to bring your application down. These range from whole operating system upgrades to hardware maintenance and the end user will never see the impact. A sysplex literally is designed to be a 24x7 operation.
You can buy other types of machines that will be more powerful, faster or do operation x better, but it is hard to find a set of machines that are as stable and reliable as a mainframe is (and process millions of transactions per second).
Also, in terms of virtualization - a single mainframe on z/Linux can host many virtual linux servers - enough that you can save a substantial amount on power costs (my org estimated 400k a year in savings in terms of power alone - if the linux servers that are hosted individually on one of our distributed networks went to virtual on a mainframe).
Re:Naming (Score:3, Informative)
Re:Imagine... (Score:3, Informative)
Except the processor... (Score:3, Informative)
Note the 'z' in eclipz. They seem to be seeking to consolidate their non-x86 offerings in terms of core component design.
Re:This might be a dumb question... (Score:3, Informative)
Translation (Score:3, Informative)
Re:Kinda slow, eh? (Score:3, Informative)
I'm a mainframer from way back and I've got the grey hair to prove it.
Re:Kinda slow, eh? (Score:3, Informative)
Yes, it IS really that easy. The cooling lines are all quick disconnect and you literally shove a module ( about the size of a typical intel box ) into an empty bay, and the system will POST, recognize and begin assigning work to another 64 processors. I have seen it with my own eyes, and it is just insanely cool!
I know a lot of /.rs are to young to remember VM / PROFS and stuff like that. VM will let you run just about any operating system as a "Guest OS" and that is some cool shit.
Re:Nah (Score:2, Informative)
Re:Kinda slow, eh? (Score:2, Informative)
Re:Imagine... (Score:2, Informative)
Re:Mainframes still around (Score:3, Informative)
Re:Kinda slow, eh? (Score:3, Informative)
Mainframes are almost like networks in a box. They're all about I/O bandwidth: moving large amounts of data from one place to another where useful work can be done on it. Individual CPUs don't have eyepopping performance because that's not how you increase the amount of work that gets done on a mainframe. You add more CPUs and attach them to the fat data pathways. If you have tasks like cryptography that might tie up CPUs, you offload it onto a co-processor. If you have tasks that require very fast individual computations rather than aggregate performance, you need a supercomputer, which is a different beast.
It's true you can design a microprocessor that looks pretty powerful compared to mainframe processors, but the trick is to find a way to keep it busy. If you are doing a non-I/O intensive task like doing an integer computation benchmark, of course it's going to look "more powerful". As soon as you do something that requires processing lots of data, then your microprocessor is spending the vast majority of its time twiddling its very muscular thumbs because you can't give it enough data to work on. Mainframes are designed so this doesn't happen.
Mainframes are not optimal for every kind of workload, but where moving data around is the bottleneck, you eventually either get a mainframe, or construct the equivalent of a mainframe yourself out of racks of servers, SAN, server virtualization and management software.