Re: Performance 2.2.13 and USB work

Björn Lundberg [off-list ref] writes:

We've been using 2.2.5 and wanted to move up to 2.2.13

So, we made a diff on all changes we've done on Malek's 2.2.5 and
applied that to 2.2.13.
We then ran a test on it and found 2.2.13 was 3 times slower than 2.2.5!

We haven't analyzed what made the difference (or in fact what changes
we've introduced to 2.2.13 with the patch, other than the reported
conflicts).

Is there anyone that experienced the same fenomena?
What might be the reason for this?

The fact that Dan's 2.2.13 uses write through cache is one thing that
will affect performance. Grep for DC_SFWT in kernel/head.S and change
the #if 0 to #if 1.

I don't know if that's the only reason, but we don't get any
performance degradation here with 2.2.13.

//Marcus
--
Signature under construction, please come back later.

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Marcus Sundberg wrote:

The fact that Dan's 2.2.13 uses write through cache is one thing that
will affect performance. Grep for DC_SFWT in kernel/head.S and change
the #if 0 to #if 1.

I did this because many people were using the standard
distribution libraries, which don't work with copyback
enabled.

For a real treat, download the linuxplanet-2.2.13 kernel
sources from MontaVista.  It includes some configuration
options that will be in the 2.3.xx sources once I get
them to compile :-).  For example, the copyback cache
and 860 Rev. B cpu6 silicon errata are configurable.


	-- Dan

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/