On Sun, 27 May 2001 wwieserat_private wrote: > Did you try to insert a usleep(1) before the writev() syscall? This > _might_ narrow down the noise as the last timer interrupt is just done when > usleep returns (On i386, the time resolution is 1/100s = 10000usec; > AFAIK usleep will not sleep shorter than time resolution.) Good idea. It eliminated a lot of noise. (Unfortunately it made tests much slower as well...my computer needed nearly an hour to make 17 x 10000 runs). The results with usleep(1), 10000 runs per a line. match. min avg max bytes 0 46451 47434 239600 1 46334 47388 156628 2 46337 47468 164785 3 46495 47521 156686 4 46581 47582 156550 5 46545 47552 166340 6 46479 47490 156573 7 46397 47393 156736 8 46522 47476 164105 9 46402 47434 156612 10 46487 47419 157070 11 46412 47446 164578 12 46498 47461 156842 13 46405 47521 156954 14 46485 47494 164273 15 46425 47425 156649 16 46397 47380 156788 --Pavel Kankovsky aka Peak [ Boycott Microsoft--http://www.vcnet.com/bms ] "Resistance is futile. Open your source code and prepare for assimilation."
This archive was generated by hypermail 2b30 : Mon May 28 2001 - 21:42:59 PDT