Bill Sommerfeld wrote:
> > It's possible the poolifying of mbufs did make the performance difference
> > smaller. It would be interesting to try it again on those machines.
> 
> You may want to try this in a non-DIAGNOSTIC kernel with the changes I
> just committed to subr_pool.c and pool.h; I measured a 1.2%
> performance improvement (on fast hardware) due to "filing off" some
> rough edges in the pool_get/pool_put code..

WOW.

With these changes, kernel without DEBUG/DIAGNOSTIC, the 3c529
does ~800KB/s on the same hw configuration.

I'm really impressed.

Amazing, I'm just plain shocked. Wow.

Jaromir
-- 
Jaromir Dolecek <jdolecek@NetBSD.org>      http://www.ics.muni.cz/~dolecek/
NetBSD - just plain best OS! -=*=- Got spare MCA cards or docs? Hand me them!