generic x86 questions - Solaris

This is a discussion on generic x86 questions - Solaris ; Hi, can anyone confirm/revise these statements (heard from a Sun Consultant): 1: a 4 CPU x86 system isnn't bootable if CPU0 fails. Never. No trick available. 2: Performance decreases significantly (up to 40%) if CPU3 fails. Not CPU 0, 1, ...

+ Reply to Thread
Results 1 to 8 of 8

Thread: generic x86 questions

  1. generic x86 questions

    Hi,

    can anyone confirm/revise these statements (heard from a Sun Consultant):

    1: a 4 CPU x86 system isnn't bootable if CPU0 fails. Never. No trick
    available.

    2: Performance decreases significantly (up to 40%) if CPU3 fails. Not
    CPU 0, 1, nor 2. Exactly CPU3. No idea why.

    3: x86 4-CPU system: large amount of RAM (~32GB) will decrease PCI
    performance up to 50%. Also no idea why.


    Thanks,

    Michael

  2. Re: generic x86 questions

    Michael Weis writes:

    >can anyone confirm/revise these statements (heard from a Sun Consultant):


    It looks like all these depend very much on the specific hardware
    platform.

    >1: a 4 CPU x86 system isnn't bootable if CPU0 fails. Never. No trick
    >available.


    I would think this depends on the hardware and the BIOS; note
    that on AMD, each CPU takes part in controlling memory and I/O so
    all I/O and memory connected to such a CPU would not work.

    On Xeon platforms, they're all connected to a separate memory controller
    and the FSB; on such platforms such a restriction needs not apply.

    It could be that legacy platform issues make it impossible to start
    with anything but CPU0.

    >2: Performance decreases significantly (up to 40%) if CPU3 fails. Not
    >CPU 0, 1, nor 2. Exactly CPU3. No idea why.


    Very much platform specific, I would think. (Is it the CPU in socket 3
    or is it the second core of the CPU in the second socket)

    >3: x86 4-CPU system: large amount of RAM (~32GB) will decrease PCI
    >performance up to 50%. Also no idea why.


    PCI performance? It is the case that in many systems filling all banks
    will cause the performance of the memory to be lower; this would impact
    overall performance, not PCI performance.

    In all, I would say that "there might be platforms which behave this way,
    but there does not seem to be a requirement that they do".

    Also, this is not a Solaris specific question; I would suggest asking
    it in comp.arch where people understand better how systems are build.

    Casper
    --
    Expressed in this posting are my opinions. They are in no way related
    to opinions held by my employer, Sun Microsystems.
    Statements on Sun products included here are not gospel and may
    be fiction rather than truth.

  3. Re: generic x86 questions

    Hi Casper,

    >
    > Also, this is not a Solaris specific question; I would suggest asking
    > it in comp.arch where people understand better how systems are build.
    >


    Thank you for your hints!
    After re-installing my system I will ask in comp.arch, that's a good idea!

    Michael


  4. Re: generic x86 questions

    On 23 Feb 2007 10:10:17 GMT, Casper H.S Dik wrote:
    > Michael Weis writes:


    >>3: x86 4-CPU system: large amount of RAM (~32GB) will decrease PCI
    >>performance up to 50%. Also no idea why.

    >
    > PCI performance? It is the case that in many systems filling all banks
    > will cause the performance of the memory to be lower; this would impact
    > overall performance, not PCI performance.
    >
    > In all, I would say that "there might be platforms which behave this way,
    > but there does not seem to be a requirement that they do".


    Indeed. A while ago, I remember that the 1st generation of Sun Opteron
    machines had this sort of limitation (the largest supported memory
    config necessitated a lower speed). I've not heard anything like that
    for the current batch of machines.

    A bientot
    Paul


  5. Re: generic x86 questions

    * Michael Weis:

    > can anyone confirm/revise these statements (heard from a Sun Consultant):
    >
    > 1: a 4 CPU x86 system isnn't bootable if CPU0 fails. Never. No trick
    > available.


    Depends a lot on the hardware design. Might be that Sun x86 systems
    can't boot with broken CPU0, but that is not necessarily true for other
    x86 systems...

    > 2: Performance decreases significantly (up to 40%) if CPU3 fails. Not
    > CPU 0, 1, nor 2. Exactly CPU3. No idea why.


    Also not generally true. On shared memory systems the performance
    decrease usually is independent on which CPU fails. On the other side,
    in NUMA configurations it's possible that loosing CPU3 causes the most
    performance penalty if most system memory and/or expansion is connected
    to them, but again this depends on the hardware design and can be true
    for every other CPU...

    > 3: x86 4-CPU system: large amount of RAM (~32GB) will decrease PCI
    > performance up to 50%. Also no idea why.


    Can't tell you about Sun machines but systems from other vendors don't
    suffer from decreased PCI performance if fitted with lots of RAM.

    Benjamin

  6. Re: generic x86 questions

    Thank you for your answers!

    Michael

  7. Re: generic x86 questions

    Michael Weis wrote:
    > Hi,
    >
    > can anyone confirm/revise these statements (heard from a Sun Consultant):
    >
    > 1: a 4 CPU x86 system isnn't bootable if CPU0 fails. Never. No trick
    > available.


    Incorrect.
    We have an ancient 4-CPU Pentium Pro server (4 x 200MHz) from Intel
    (prototype) running Solaris 10 6/06. Once upon a time we lost a board
    and with it, two CPUs automatically went buh-bye. The board had been
    taken out, and the system was booted up with two CPUs until we got the
    necessary parts.

    Also, Solaris on i86pc can now use the `psrset` command to switch CPUs
    offline or online on the fly. However, I believe this will only work on
    ACPI capable i86pc systems.

    > 2: Performance decreases significantly (up to 40%) if CPU3 fails. Not
    > CPU 0, 1, nor 2. Exactly CPU3. No idea why.


    Makes no sense.

    > 3: x86 4-CPU system: large amount of RAM (~32GB) will decrease PCI
    > performance up to 50%. Also no idea why.


    Also makes no sense without hard data to back it up and a detailed
    kstat, lockstat, vmstat and iostat analysis. Or DTrace, for that matter.

  8. Re: generic x86 questions

    Michael Weis wrote:
    > Hi,
    >
    > can anyone confirm/revise these statements (heard from a Sun Consultant):
    >
    > 1: a 4 CPU x86 system isnn't bootable if CPU0 fails. Never. No trick
    > available.


    Just to clarify: I believe that the board that went buh-bye had CPUs 0
    and 1 on her.

    Whether the system can boot or not after such a failure from one of the
    other CPUs depends solely on how intelligent the BIOS and how
    redundantly designed the hardware is.

+ Reply to Thread