Subject: Re: Strange libc shared vs. unshared performance
To: None <port-sparc@netbsd.org>
From: Vareck Bostrom <bostrov@lava.cs.orst.edu>
List: port-sparc
Date: 11/16/1994 02:21:23
In message <199411160851.TAA05847@anarres.mame.mu.oz.au>, Daniel Carosone write
s:
>
>For the sake of curiosity, I took the dhry executables I built on
>sunos for comparison, and ran *those* on my netbsd/sparc.
>
>They give basically the same result, regardless of static or dynamic
>linking, and that result is the same as the netbsd -static one.
>
>Just another datapoint..

My friend (kelly@supra.com) just put NetBSD 1.0 on his sparcstation 
last night, with the help of thorpej and myself. One of the first things
he noticed was how terribly SLOW vi (/usr/bin/vi) started up. He compared
it to a 80386 40MHz BSDI box, and the BSDI vi was clearly faster. 

I recompiled vi -static, and now it shows normal performance. 

There is a definite problem somewhere here. We're recompiling all our
commonly used programs (login, w, top, vi, emacs, etc, etc,) and 
already the machines are responding MUCH MUCH faster and seem like real
SPARCstations now, rather than rejects from the Sun-3 era. 

It's not just dynamic linking, it's dynamic linking the shared libc 
on the SPARC NetBSD port. NOTHING else shows this behavior. Not even
solaris 2.3.

Any clues as to how to fix this, guys? I would consider this to be a 
serious problem. Things run, but so slowly as to reduce your 25 MHz SPARC
to a 10-16 MHz 68020 in speed, if that.