On Tue, Jan 07, 2003 at 12:14:31PM +1100, matthew green wrote:
>    
>    Note that top says 50% of the CPU is idle, which is consistent with the
>    fact that both processes are running on CPU 0. I'm running the last
>    current userland from releng.netbsd.org
> 
> 
> right, you need to patch locore.s about line 4275 and make the
> #if 0 an #if 1....

OK, much nicer now:
load averages:  1.81,  0.76,  0.31                                     21:48:08
32 processes:  1 runnable, 29 sleeping, 2 on processor
CPU states: 40.2% user,  0.0% nice, 59.3% system,  0.5% interrupt,  0.0% idle
Memory: 12M Act, 260K Wired, 1860K Exec, 3136K File, 41M Free
Swap: 195M Total, 195M Free

  PID USERNAME PRI NICE   SIZE   RES STATE      TIME   WCPU    CPU COMMAND
  212 bouyer    64    0   720K 1328K CPU/1      1:28 99.04% 98.34% tcsh
  207 bouyer    63    0   720K 1328K RUN/1      1:24 97.56% 97.02% tcsh
    5 root      18    0     0K   15M syncer/0    ???  0.05%  0.05% [ioflush]
  216 bouyer    28    0   204K  920K CPU/0      0:00  0.05%  0.05% top


The hight system usage looks weird for a CPU bound process, but I think
I've seen this on sparc for ages. The way system ws user CPU time is
accounted on sparc may be wrong.

I'm now going to compile a kernel -j10 and see how it performs.

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
     NetBSD: 23 ans d'experience feront toujours la difference
--