Re: comparison of the NOVM version

Andres Kroonmaa wrote:
>
> > I've been running an experiment to try to quantify the differences
> > between the regular Squid and the NOVM version. The results are
> > at http://squid.nlanr.net/Squid/Devel/NOVM-compare/.
> >
> > Comments would be appreciated, as well as any other results comparing
> > the two versions.
> >
> > Duane W.
>
> This is very interesting. I'm surprised that NOVM performs that close
> to VM version, I'd expect it to be much slower.
>
> Guess: NOVM is slightly faster on HIT's because there is no delay of
> objects being swapped in into VM initially.
>
> Then again, if pushed by zillions of small MISS requests I think NOVM
> would be slower somewhat as each fileopen adds additional time to
> request service times.
Just a simple test:

linux 2.0.28, Pentium 75, 32M of ram:

> time ./open-test
0.020u 0.890s 0:01.66 54.8% 0+0k 0+0io 66pf+0w
> cat open-test.c

#include <stdio.h>

main(){
int a;
int b;

for (a=0;a <10000;a++){
        b = open("temp.a","rw+");
        close(b);
        }
}

10000 open and closes of a file takes just over 1.5 seconds (and yes, gcc
hasn't optimised the system calls out of existance, as an strace shows
the calls.

I really don't think that this is a limiting factor... of course reading
10000 places in ram is a lot faster... but the time difference is
really worth ignoring.

I know that opening and closing files in random places on the disk is a
better idea, but it'll take to long to write the code ;)

        Oskar

=============================================
'experience made art, but inexperience luck.'
=============================================
Received on Tue Jul 29 2003 - 13:15:41 MDT