Finally managing to compile the 2.3.8 kernel (I forgot to exclude minix
support), I also ran a few benchmarks vs. 2.2.{9,10}. Instead of such
CPU intensive tasks a compiling the kernel, I decided to grep my 1Gig
/usr and to updatedb. Here are the results:
Kernel 2.2.9:
# du -s /usr
1400546 /usr
# time updatedb
real 2m15.953s
user 0m11.210s
sys 0m18.430s
# time updatedb
<n/a>
# time updatedb
real 0m30.105s
user 0m10.910s
sys 0m17.750s
Kernel 2.2.10:
# time updatedb
real 2m20.636s
user 0m11.330s
sys 0m18.350s
# time updatedb
real 0m29.365s
user 0m10.990s
sys 0m17.670s
# time updatedb
real 0m29.778s
user 0m10.940s
sys 0m17.820s
# find /usr -type f -print |split
# time for i in x*; do cat $i|xargs grep Linus > /dev/null; echo $i
done; done
real 4m31.633s
user 0m15.420s
sys 0m49.510s
Kernel 2.3.8:
# time updatedb
real 2m24.827s
user 0m11.540s
sys 0m19.420s
# time updatedb
real 0m30.744s
user 0m11.330s
sys 0m19.160s
# time updatedb
real 0m30.776s
user 0m11.290s
sys 0m19.300s
# find /usr -type f -print |split
# time for i in x*; do cat $i|xargs grep Linus > /dev/null; echo $i
done; done
real 4m59.729s
user 0m15.060s
sys 0m52.150s
As you can see, the fs performance of 2.3.8 w.r.t. to find/grep tasks
never challanges 2.2.x performance. The second and third updatedb runs
are almost entirely from buffers, i.e. there was very little disk
activity there.
System specs:
AMD K6-2-300, 128M RAM, 2x4.5G IBM DDRS U2W SCSI @ Symbios Logic U-SCSI
/home,/opt,/usr: raid0
/tmp,/var: linear append
/: just normal
The kernels were almost identically configured, except:
2.2.10 was with framebuffer, others not
2.3.8 was w/o minixfs, others with
So what would be a (real life) test that sees 2.3.8 *gaining*
performance over 2.2.x?
Marc
Marc
-
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.tux.org/lkml/