> were these iozone runs done with a filesystem or on a raw device?

It's FFS.

> if it was with a filesystem, you could also try a run with the performence
> enhancements I've been working on:
> 
> ftp://ftp.netbsd.org/pub/NetBSD/misc/chs/ubc-perf/
> 
> grab the latest diff, put the new genfs_node.h file in sys/miscfs/genfs,
> rebuild.  this improves filesystem performance dramatically on old, slow
> machines.

I've compiled and measured.

K48     1.5K compiled at 2000-11-21
X1      1.5X                 -08-22
CHS0	1.5X                 -08-23

          K48           X1              CHS0
W sec.  36.38           72.98           71.43
R sec.  29.79           38.27           35.10

W B/s   1,152,914         553,411         587,190
R B/s   1,407,957       1,083,799       1,194,958

CHS0 performance shows very little fluctuation, less than 0.7% this
time.

It's known that the combination of 25MHz R3000A and LPS540S yields
1,700,000+ B/s on raw write and 2,300,000+ B/s on raw read.  Raw I/O
performance figures are consistent between K48 and CHS0, despite of
modifications of MI NCR53C9x SCSI driver and new SCSIPI layer.

According to the output of vmstat 1, the kernel emits 150+ page faults
every second.

Tohru Nishimura