Re: 2.6.8.1-mm3

From: Jesse Barnes (jbarnes_at_engr.sgi.com)
Date: 08/20/04

  • Next message: Jean-Luc Cooke: "Re: [RFC] enhanced version of net_random()" 
    To: Andrew Morton <akpm@osdl.org>
Date:	Fri, 20 Aug 2004 15:56:10 -0400

    On Friday, August 20, 2004 2:55 pm, Andrew Morton wrote:
    > > Here's the output part way through a kernbench run:
    >
    > (This doesn't sound like the sort of workload which people would buy an
    > Altix for?)

    No, but other people are working on that stuff (e.g. Christoph's page faulting
    code). That said, good multiuser performance is important on big boxes too.

    > > 27081955 ia64_pal_call_static 141051.8490
    > > 3175264 ia64_load_scratch_fpregs 49613.5000
    > > 3166434 ia64_save_scratch_fpregs 49475.5312
    >
    > What do the above three mean?

    ia64_pal_call_static is part of the idle path. Puts the CPU into low power
    state. The save and restore fp regs I think are due to the fact that integer
    multiply uses fp regs, David? They're also preserved across PROM calls, but
    we shouldn't be preserving them unless necessary.

    > > 1603765 ia64_spinlock_contention 16705.8854
    >
    > That's 0.04% of total non-idle CPU time. This seems wrong.

    The time between resetting the profiling buffer and collecting the profile was
    small, and was before kernbench had started 2048 threads, so most of the CPUs
    were idle. I can do a longer run now that the profiling stuff seems to work.

    Thanks,
    Jesse
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

  • Next message: Jean-Luc Cooke: "Re: [RFC] enhanced version of net_random()"

    Relevant Pages

    • Re: Memory allocation performance
      ... -- "high resolution profiling" is of increasingly limited utility with modern CPUs, where even a high frequency timer won't run very often. ... I am now using Pentium4 Prescott CPU with HTT enabled in BIOS, but kernel built without SMP to simplify profiling. ...
      (freebsd-hackers)
    • Re: Memory allocation performance
      ... -- "high resolution profiling" is of increasingly limited utility with modern CPUs, where even a high frequency timer won't run very often. ... I am now using Pentium4 Prescott CPU with HTT enabled in BIOS, but kernel built without SMP to simplify profiling. ...
      (freebsd-performance)
    • Re: sched_yield behavior
      ... >> CPU, to give to other tasks a chance to run. ... >> higher priority task, and the only of its priority list). ... > the exact nature of the real time constraint? ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: [patch] Real-Time Preemption, -RT-2.6.10-rc1-mm2-V0.7.1
      ... >X should be scheduled on the other CPU just fine. ... Only per-CPU kernel ... those should schedule fine too. ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)
    • Re: [PATCH] implement nice support across physical cpus on SMP
      ... PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND ... send the line "unsubscribe linux-kernel" in ...
      (Linux-Kernel)