From: Henry N. <Hen...@Ar...> - 2008-03-29 15:22:24
|
Hello, Óscar Fuentes wrote: > Henry Nestler <Hen...@Ar...> writes: > >> Óscar Fuentes wrote: >>> Henry Nestler <Hen...@Ar...> writes: >>> >>>>> The test suite starts a process more than 1000 times, pipes the output >>>>> into the controlling process, etc. My experience shows that GNU/Linux is >>>>> more effective than Windows for this sort of tasks. >>>> Suggest from you, here is a test script, that only eceute some >>>> processes, without disk io: >>>> #!/bin/bash >>>> >>>> loop=10000 >>>> while test $loop -gt 0 >>>> do >>>> x=`uname -r` >>>> loop=$(($loop - 1)) >>>> done >>>> >>>> Tested with Intel(R) Pentium(R) 4 Mobile CPU 1.80GHz >>>> 22.9s native, kernel 2.6.17 >>>> 89.7s coLinux 0.8.0, kernel 2.6.22.18 scnapshot-20080329 http://www.colinux.org/snapshots/ is significant faster now: 30.1s coLinux 0.8.0-20080329, kernel 2.6.22.18 Round about 500.000 OS switches in this test script are optimised with a better handling of page allocation. The gust call to the host never, if the page is always allocated. The changes on guest side would also speed up some other programms. Typically all programms with high numbers of memory alloc/free runs faster. The other test http://ftp.gnu.org/pub/gnu/make/make-3.81.tar.gz is now: configure make native: 28s 15s coLinux-0.7.3: 63s (41%) 28s (53%) coLinux-20080329: 31s (90%) 19s (79%) -- Henry N. |