lkml.org 
[lkml]   [2000]   [Oct]   [31]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: 2.2.18Pre Lan Performance Rocks!


    "Jeff V. Merkey" wrote:
    >
    > Pavel Machek wrote:
    > >
    > > Hi!
    > >
    > > > > > This is putrid. NetWare does 353,00,000/second on a Xenon, pumping out
    > > > > > gobs of packets in between them. MANOS does 857,000,000/second. This
    > > > > > is terrible. No wonder it's so f_cking slow!!!
    > > >
    > > > And please check your numbers, 857 million
    > > > > context switches per second means that on a 1 GHZ CPU you do one context
    > > > > switch per 1.16 clock cycles. Wow!
    > > >
    > > > Excuse me, 857,000,000 instructions executed and 460,000,000 context
    > > > switches
    > > > a second -- on a PII system at 350 Mhz. It's due to AGI
    > > > optimization.
    > >
    > > That's more than one context switch per clock. I do not think
    > > so. Really go and check those numbers.
    >
    > Pavel, The optimization exploits the multiple piplines in Intel's
    > processors,
    > and yes, it does execute more than one instruction per clock, it's
    > optimized
    > to execute in the processors parallel pipelines. The EMON numbers are
    > accurate,
    > and you can download the kernel and verify for yourself. These types of
    > optimizations
    > are possible when people have acccess to Intel Red Cover documents, then
    > you
    > get to know just how Intel's internal architectures are affected by
    > different coding optimizations.
    >
    > Jeff

    There's also another optimization in this kernel that allows it to
    achieve greater than
    100% scaling per processor by using a strong affinity algorithm (I hold
    the patent
    on this algorithimn, and by posting code based on it under the GPL, I
    have released
    it to the general public).

    It relies on an anomoly in the design of Intel's cache controllers, and
    with memory based
    applications, I can get 120% scaling per procesoor by jugling the
    working set of
    executable code cached accros each processor. There's sample code with
    this kernel
    you can use to verify....

    :-)

    Jeff


    > > Pavel
    > > --
    > > I'm pavel@ucw.cz. "In my country we have almost anarchy and I don't care."
    > > Panos Katsaloulis describing me w.r.t. patents at discuss@linmodems.org
    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2005-03-22 12:45    [W:0.025 / U:1.908 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site