I copied all 600MB worth of files from a IDE CD (24x) to /dev/null
using standard cp utility and cp imitation using sendfile() system
call instead of read()/write().
Results are as follows:
sendfile cp:
0.73user 34.23sys 3:42.18real 15%CPU (170120major+4636minor)pagefaults 25swaps
standard cp:
3.07user 156.15sys 3:44.34real 70%CPU (172868major+4878minor)pagefaults 25swaps
Wallclock times are equal since we're bound with the speed of CD-ROM
drive, but look at the CPU utilization. We have a clear winner, here.
What bothers me is that even though sendfile uses much less CPU time,
machine in fact becomes unusable while a program using it is running.
When using standard cp you barely notice slowdown, but sendfile()
definitely kills interactive feel.
Is there maybe a subtle bug in sendfile() implementation?
Or could it be that kernel incorectly measures and reports CPU usage?
Something else?
Pavel, if I recall correctly, you once had a patch you used to improve
kernel fairness in reporting CPU usage.
Could you be so kind and send me a copy, and spare me losing myself in
the archives. Also, I would like to have a recent version, for
2.1.115, if possible.
Thanks!
-- Posted by Zlatko Calusic E-mail: <Zlatko.Calusic@CARNet.hr> --------------------------------------------------------------------- It really bothers me when people cut me o...- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html