Re: ten thousand small processes

From: D. J. Bernstein (djb_at_cr.yp.to)
Date: 06/26/03

  • Next message: D. J. Bernstein: "The dangers of replacing malloc()" 
    Date: 26 Jun 2003 02:50:29 -0000
To: freebsd-performance@freebsd.org

    Jon Mini writes:
    > I'm sorry, but you are way off here. First of all, caches are *much
    > larger* than the size of the processes you are talking about.

    I'm sorry, but you are being misled by a naive model of CPU performance.
    On a typical Pentium in our department, the following program becomes
    three times faster when SPACING is changed from 4096 to 128:

       #define SPACING 4096
       char data[8 * SPACING];
       main()
       {
         int i;
         for (i = 0;i < 10000000;++i) {
           data[0] = data[SPACING];
           data[2 * SPACING] = data[3 * SPACING];
           data[4 * SPACING] = data[5 * SPACING];
           data[6 * SPACING] = data[7 * SPACING];
         }
       }

    >From an asm programmer's perspective, when FreeBSD decides to spread a
    small program's variables between

       * the beginning of a data page,
       * the beginning of a bss page,
       * the beginning of a malloc mmap page,
       * the beginning of a heap page,
       * the beginning of the next heap page,
       * the beginning of yet another heap page,

    et cetera, it is actively trying (with varying degrees of success) to
    damage cache performance in exactly the same way that this program does.

    ---D. J. Bernstein, Associate Professor, Department of Mathematics,
    Statistics, and Computer Science, University of Illinois at Chicago
    _______________________________________________
    freebsd-performance@freebsd.org mailing list
    http://lists.freebsd.org/mailman/listinfo/freebsd-performance
    To unsubscribe, send any mail to "freebsd-performance-unsubscribe@freebsd.org"

  • Next message: D. J. Bernstein: "The dangers of replacing malloc()"