Numerical algorithm running slow on P4 system. - Questions

This is a discussion on Numerical algorithm running slow on P4 system. - Questions ; I have an old 500MHz Pentium 3 running SuSE Linux 7.1 (kernel 2.4.0). I needed to speed up the execution of a numerical algorithm running in scilab, so I put together a 3GHz P4 XEON APIC v16 on a Shuttle ...

+ Reply to Thread
Results 1 to 2 of 2

Thread: Numerical algorithm running slow on P4 system.

  1. Numerical algorithm running slow on P4 system.

    I have an old 500MHz Pentium 3 running SuSE Linux 7.1 (kernel 2.4.0). I
    needed to speed up the execution of a numerical algorithm running in scilab,
    so I put together a 3GHz P4 XEON APIC v16 on a Shuttle FB61 main board with
    SuSE Linux 9.0 (kernel 2.4.21-99 SMP4G). To my surprise, the *old* P3 system
    runs the scilab program *faster* than the new P4. The vmstat command gives
    the following CPU information when the scilab program is iterating.
    ----- cpu -----
    us sy id wa
    P3 82 18 0
    P4 8 43 49 0

    This seems to explain the slowness, the P4 should give 6 times the CPU
    cycles of the P3, but the application (us) is only getting 8% of these
    cycles and 0.08*6=0.48<<0.82 . Can anyone suggest how I can use the
    power of the CPU more effectively?

    As an aside, I am able to boot the P3 under Windows
    NT 4 as well and it turns out that the P3 Linux system goes faster than
    NT on the scilab program by a factor of 2.8, so I was really hoping that the
    new Linux P4 would be really quick.

    The scilab program does display an image on each iteration, but each iteration
    takes about 10sec on the Linux P3 so graphical performance is not important.

  2. Re: Numerical algorithm running slow on P4 system.

    stebla@ntlworld.com (Stephen Blake) wrote:

    > I have an old 500MHz Pentium 3 running SuSE Linux 7.1 (kernel 2.4.0). I
    > needed to speed up the execution of a numerical algorithm running in scilab,
    > so I put together a 3GHz P4 XEON APIC v16 on a Shuttle FB61 main board with
    > SuSE Linux 9.0 (kernel 2.4.21-99 SMP4G). To my surprise, the *old* P3 system
    > runs the scilab program *faster* than the new P4. The vmstat command gives
    > the following CPU information when the scilab program is iterating.
    > ----- cpu -----
    > us sy id wa
    > P3 82 18 0
    > P4 8 43 49 0


    OK, I think the real problem is this: why should a very fast
    system devote so much cpu to system? If you execute top, what
    processes float to the top? What does top or ps aux show when
    you're not running scilab?

    With such a fast system, the cpu devoted to system should be
    very small. Your scilab process is being starved out; this is wrong.

    Frankly, it looks to me like a scheduler bug. My advice is to
    simplify the problem to try to better describe it, and then take
    it up with kernel developers.

    For what it's worth,
    Robert Dodier

+ Reply to Thread