QLin=F8nut= screen shots 10770
 |
One last bit. With the work laptop I took to Ireland, I ran the process tests. The hardware is a P-III Coppermine at 700MHz, running Windows 2000 Server and Knoppix 3.8. The results are interesting:
For Windows: Ray's windows proc test, 15 iterations. min: 2725 max: 123997 avg: 11234 Ray's windows proc test, 1500000 iterations. min: 1333 max: 12521585 avg: 2070 Ray's windows proc test, 1500000 iterations. min: 1329 max: 7327082 avg: 2064 Ray's windows proc test, 15000000 iterations. min: 1319 max: 38862704 avg: 2126 Ray's windows proc test, 15000000 iterations. min: 1320 max: 39580152 avg: 2108
Ray's tweaked version of OK's double-event thread test, 1500000 iterations. min: 989 max: 5588406 1705 (Average CPU cycles) Ray's tweaked version of OK's double-event thread test, 15000000 iterations. min: 984 max: 39533583 1722 (Average CPU cycles) Ray's tweaked version of OK's double-event thread test, 15000000 iterations. min: 984 max: 40261864 1711 (Average CPU cycles) Ray's tweaked version of OK's double-event thread test, 15000000 iterations. min: 996 max: 40688158 1749 (Average CPU cycles) Ray's tweaked version of OK's double-event thread test, 15000000 iterations. min: 984 max: 40993109 1718 (Average CPU cycles)
For Linux (note that it was compiled for SMP, though):
Linux Knoppix 2.6.11 #6 SMP Wed Feb 23 09:15:46 CET 2005 i686 GNU-Linux NPTL 0.60
And here's the Windows thread test. I've tweaked it just a little from OK's original code, but I do not believe I have changed its fundamental behavior in any significant way. ----- oktest.cpp ----- #include "stdafx.h...
Ray's first process-switch test, 1500000 iterations. min: 3083 max: 10403532, avg: 3425. Ray's first process-switch test, 1500000 iterations. min: 3247 max: 10234796, avg: 3443. Ray's first process-switch test, 1500000 iterations. min: 3252 max: 3893280, avg: 3421. Ray's first process-switch test, 1500000 iterations. min: 3071 max: 10370387, avg: 3441. Ray's first process-switch test, 1500000 iterations. min: 3052 max: 10732077, avg: 3422.
Here's the Unix version: ------ uproctest.c; compile with "gcc -O2 uproctest.c -o uproctest ------ #define ITER 1500000 typedef union semun { int val; struct semidds *buf; ushort * array; } SEMARG...
Ray's unix thread test, 1500000 iterations... min: 2456 max: 1102414 avg: 2687. Ray's unix thread test, 1500000 iterations... min: 2470 max: 1117573 avg: 2677. Ray's unix thread test, 1500000 iterations... min: 2460 max: 1784813 avg: 268 plus 16. Ray's unix thread test, 1500000 iterations... min: 2460 max: 1020166 avg: 268 plus 11. Ray's unix thread test, 1500000 iterations... min: 2470 max: 1785566 avg: 2676.
Gotta admit, with the current code, Windows does come out better than Linux in everything except maximum switch times. Linux's best and average took longer, but the worst case for Linux was better than the worst case for Windows. Still don't have a good uniprocessor comparison on the same hardware, though, cause all the Linux kernels I've been able to run on Windows machines have been SMP kernels. Interesting that Knoppix is compiled SMP...
-- Sincerely,
Here are my results, on a dual 2.1GHz ("2.6Ghz Intel equivalent") Athlon MP system. Both tests were run in SMP mode, there wasn't time to test with one CPU disabled. On this machine, XP was...
Ray Ingles (313) 227-2317
"If decisive but mbuttive killing actually stopped the other side, Israel would be as peaceful as Omaha by now." - elmegil