On Wed, Apr 25, 2001 at 02:40:24AM +1000, Simon Burge wrote:
> [...]
> I notice that people haven't been showing latency results (from
> lat_pipe) - but it's good to see those numbers better too on
> these two platforms.

I didn't notice lat_pipe :)
Hree are updated results:
Vax (MicroVAX 3300, KA640 CPU):
text    data    bss     dec     hex     filename
813688  19256   158120  991064  f1f58   netbsd-newpipe
809764  19224   157852  986840  f0ed8   netbsd-oldpipe 
old pipes: Pipe bandwidth: 0.85 MB/sec Pipe latency: 3099.0735 microseconds
new pipes: Pipe bandwidth: 8.85 MB/sec Pipe latency: 3006.7038 microseconds

Next68k (68040@25Mhz):
text    data    bss     dec     hex     filename
960176  44964   147856  1152996 1197e4  netbsd-newpipe
955496  44932   147588  1148016 118470  netbsd-oldpipe
old pipes: Pipe bandwidth: 1.12 MB/sec Pipe latency: 1151.4129 microseconds
new pipes: Pipe bandwidth: 3.59 MB/sec Pipe latency:  702.0503 microseconds

macppc (603ev@160Mhz)
text    data    bss     dec     hex     filename
1872008 34364   243436  2149808 20cdb0  netbsd-newpipe
1865772 34328   243196  2143296 20b440  netbsd-oldpipe
old pipes: Pipe bandwidth: 14.92 MB/sec Pipe latency: 132.8331 microseconds
new pipes: Pipe bandwidth: 16.81 MB/sec Pipe latency:  60.9367 microseconds

So: no much changes in latency on vax; the delay may be caused by something
else (context switching maybe?)
Even on the macppc where the bandwidth improvement isn't important, the
latency is cut by half.

--
Manuel Bouyer, LIP6, Universite Paris VI.           Manuel.Bouyer@lip6.fr
--