Announcement

Collapse
No announcement yet.

Linux 3.20 To Fully Support The IBM z13

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Linux 3.20 To Fully Support The IBM z13

    Phoronix: Linux 3.20 To Fully Support The IBM z13

    Last month IBM announced the z13 micro-processor for their z13 mainframe computers. IBM claims the z13 is the "world's fastest microprocessor" and now with Linux 3.20 there's full support...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Pretty much all the online descriptions of z13 talk about "141 cores" but I think it's actually 141 processors or 1128 cores.
    Test signature

    Comment


    • #3
      Originally posted by bridgman View Post
      Pretty much all the online descriptions of z13 talk about "141 cores" but I think it's actually 141 processors or 1128 cores.
      Dont think so. Mainframes are not so great as much they are expensive. Real performance king are POWER processors, mainframe is mostly for those who like buying bridges.

      Comment


      • #4
        Yep, I think you're right about core count (24 chips @ average of 7 cores per chip = 168 cores, with some held back for spares etc..). There are a lot more processors/cores than that in the z13 but mostly used for IO.

        AFAIK the z13 and POWER8 share a lot of tech -- not just fab process, it's the same design team and they re-use blocks & elements between the Z & POWER families so it's more like "same chip with different instruction set". POWER systems tend to have more cores and less IO focus while Z systems tend to have more IO bandwidth relative to cores.

        The L4 cache is interesting, almost a GB of wide EDRAM per processor drawer.
        Last edited by bridgman; 11 February 2015, 01:12 PM.
        Test signature

        Comment


        • #5
          ..but does it run (crysis) PTS?

          Comment


          • #6
            Yes, same team makes it, but mainframe is 50 year old architercture. They need to keep backward compatibility and making such a harware is expensive and crufty.

            IBM has eClipz strategy since 90s, to converge all their systems on one processor line. They already merged IBM i and AIX hardware to Power Systems, system x is sold, next target is to put z/OS on POWER hardware without breaking binary compatibility for aplications. Probably by using modified TIMI from IBM i.

            Comment


            • #7
              As someone who missed the mainframe's golden age, here is something I don't understand and momentarily I can't dedicate time to researching this.

              I always associated mainframes with a very monolithic architecture. Aggregating a lot of performance in a single machine in order to scale up. What is the point of running virtual machines on a mainframe?
              As long a VM does not exceed the resources offered by the host a VM can run on any machine. What is the point of running very many VMs on a single very powerful machine instead of running very many VMs on several servers?

              Comment


              • #8
                Originally posted by alexvoda View Post
                As someone who missed the mainframe's golden age, here is something I don't understand and momentarily I can't dedicate time to researching this.

                I always associated mainframes with a very monolithic architecture. Aggregating a lot of performance in a single machine in order to scale up. What is the point of running virtual machines on a mainframe?
                As long a VM does not exceed the resources offered by the host a VM can run on any machine. What is the point of running very many VMs on a single very powerful machine instead of running very many VMs on several servers?
                The issue comes down to lower latency on the side of single large systems vs higher aggregate bandwidth of many small systems.

                Comment


                • #9
                  ... and fault tolerance. The "z" in z13 is for zero downtime.

                  I've seen mainframes with smoke & flame coming out of them but still running. The computer operator threw a chair at someone and missed, and the cabinets were open for maintenance, so the chair legs poked holes in the core memory planes and ripped out a bunch of wires. The system shut down power to the affected cabinets, requeued the affected jobs to restart from the last checkpoint, placed a service call and advised the operator what it had done via screen and printer.

                  I imagine you would have a lot of independent programs running on VMs but all talking to a big honkin' DBMS.

                  Remember Seymour Cray's question about what you would prefer for ploughing a field -- 2 strong oxen or 1024 chickens. Sounds like a good Mythbusters episode.
                  Last edited by bridgman; 11 February 2015, 05:02 PM.
                  Test signature

                  Comment


                  • #10
                    Get off my lawn!

                    Originally posted by alexvoda View Post
                    I always associated mainframes with a very monolithic architecture. Aggregating a lot of performance in a single machine in order to scale up. What is the point of running virtual machines on a mainframe?
                    People don't understand that Mainframes are for saving money for big data-based organizations.

                    The point of Mainframes isn't processing power, it's I/O speed. Nothing comes close: the architecture is designed from the ground up without any of the multithreading bottlenecks you get in more "modern" tech. A single Mainframe can handle dozens of I/O controllers (filling a whole room) that can power hundreds of terminals. This is significantly cheaper in terms of hardware and maintenance fees over a fleet of desktop PCs or even network-based terminal solutions (Unix, Windows server, VAX in the past). Mainframes prove again and again that newer isn't necessarily better.

                    Mainframes also have very good backup solutions, based on cheap magnetic tape that you just have to store in a library somewhere. The terrific I/O throughput means that you can backup continuously while in full operation.

                    Surprisingly, Mainframes started getting expensive as the PC-based competition ramped up: there were just not enough skilled Mainframe operators and programmers, and they were expensive. So, IBM has been slowly moving Mainframe to contemporary technologies, such as Linux and Java. It's been a breath of fresh air in the industry as it's allowed a new generation of programmers entry into this world. Well, the old-timers also complain that these people are coming in without really understand the Mainframe world and its culture, and thus make bad software.

                    Still ... times are changing. If you're already running Java on your Mainframe, you might be tempted to treat it as a regular big old server, and do crazy things like trying to run Hadoop on your db/2 data... IBM's customers for the first time are pushing for Mainframes that could actually be decent Linux servers. That's the background behind the recent upgrade to Mainframe's processing power, and the noise about making the Mainframe more attractive for "cloud" computing.

                    If you're not running a Mainframe already, you won't care about any of this: you'll be running a PC-architecture- (or, recently ARM-) based grid (or private cloud) in your data center using cheap, easy replacable parts, and you've invested millions (billions!) in distributed software architectures.

                    (I'm a certified Mainframe tech from the olden days)

                    Comment

                    Working...
                    X