Subject: Re: whee, it works.. sort of
To: None <,>
From: Toru Nishimura <>
List: port-pmax
Date: 03/25/2000 14:51:21
Kevin Schoedel reported NetBSD/pmax is runing slowly;

> I've had the feeling that something is slower than it ought to be.
>     [ ... snip ... ]
> My main home box is a 32M 5000/133 running 1.4.1; /, /usr, and home
> directories on separate spindles.
>     [ ... snip ... ]
> Slowness is most apparent when running scrips. For instance, my .profile,
> when run under near-ideal conditions (idle box, when it's just been run,
> and logging in induces no disk activity whatsoever), takes 30 seconds. On
> the 'nearest' other box I have, a 40MHz 32M SparcStation 10, the same
> .profile takes 3 seconds. Granted, the SS10 is a little faster, but it
> certainly isn't *ten times* faster, and Solaris 2.7 isn't exactly lean
> and mean.

I ran several measurements using the script he gave me, and concluded
that NetBSD/pmax actually runs very slowly for complex shell scripts.

I got dramatical performance differences between /bin/sh and /bin/ksh
for login shell, yet s-l-o-w-e-r than other ports.  The result also
endorses that NetBSD/pmax spends too much time for fork/exec sequences.
This performance issue _must be_ analyzed and fixed.  I'd like to
start looking at pmap.c, TLB or cache dynamisms. 

Tohru Nishimura