9fans archive / 1994 / 04 / 5 /    prev next

From: forsyth@min... forsyth@min...
Subject: fork test
Date: Sun, 10 Apr 1994 16:42:52 -0400

i've just tried the 10,000 fork test on a Sparc Classic

unix
	156.1 real	136.9 sys	13.5 user	dynamically linked
	 58.7 real	 50.4 sys	 3.2 user	statically linked, dynamically paged executable
	 58.8 real	 51.1 sys	 2.8 user	statically linked, not dynamically paged image (-n)

plan 9
	 26.9 real	 20.4 sys	 0.8 user	no need to choose

the sparc classic has tiny caches, and a sleazy implementation of the Reference MMU
(TLB context flushes are implemented as complete TLB flushes).  i might revise the MMU code,
but i don't think it will help this test on this model of sparc.  memmove might
be made faster, though; i haven't checked yet.  i haven't got figures for the ss/10,
since that hardware is different again.