Re: Very poor performance with 2.6.4

[Andreas Hartmann]

Chris Mason wrote:
> On Mon, 2004-03-29 at 01:16, Andreas Hartmann wrote:
> > Andrew Morton wrote:
> > > Andreas Hartmann <andihartmann@xxxxxxxxxx> wrote:
> > >>
> > >> I tested kernel 2.6.4. While compiling kdelibs and kdebase, I felt, that
> > >>  kernel 2.6 seems to be slower than 2.4.25.
> > >> 
> > >>  So I did some tests to compare the performance directly. Therefore I
> > >>  rebooted for everey test in init 2 (no X).
> > >> 
> > >>  I locally compiled 2.6.5rc2 3 times under 2.6.4 and under 2.4.25 on a
> > >>  reiserfs LVM partition, which resides onto a IDE HD (using DMA) and got
> > >>  the following result:
> > >> 
> > >>  In the middle, compiling under kernel 2.6.4 tooks 9.3% more real time than
> > >>  under 2.4.25.
> > >>  The user-processortime is about the same, but the system-processortime is
> > >>  under 2.6.4 32.9% higher than under 2.4.25.
> > > 
> > > Try mounting your reiserfs filesystems with the `-o nolargeio=1' option.
> >
> > This didn't help.
> >
> > > 
> > > If that doesn't help, please run a comparative kernel profile.  See
> > > Documentation/basic_profiling.txt.
> >
> > I'll do this next.
>
> You might also want to try 2.6.5-rc2 which has a set of reiserfs fixes
> from 2.4.x.  I'm hoping those will clean things up for you.

Ok, here is the result for 2.6.5-rc2 (3 times middle, with preemption), 
compared to 2.4.25.

The result is, the performancedifference to 2.4.25 is the same as for 
2.6.4 as described above.
Nearly means: The real processing time is about 1% faster than under 
2.6.4, but 8.3% slower than with 2.4.25. The system-processortime is 1.2% 
faster than under 2.6.4 but 31,7% more than under 2.4.25. The times for 
the user-processortime is unchanged.

But I'm not shure if these values are really significant, because the 
values for the real time meassured each try differ a lot under 2.6.

For example 2.6.5rc2:
between 9.07 min and 8.37 min for real time.
Under 2.4.25, the differences are a lot of smaller: between 8.06 min and 
8.15 min for real time.
The values for user and system time are nearly constant with 6.49 min and 
36 sec (kernel 2.6) and 6.43 min / 27 sec for 2.4.25.


Regards,
Andreas Hartmann
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/