Re: KROC & occam on Linux

["J.Moores" <jm40@xxxxxxxxx>]

Whoops - the kernel linked to these benchmarks is a debugging build!  It
should not be used for benchmarks (I've added some new figures with the
debugging off and optimisation on.)

On Tue, 7 Jul 1998, A E Lawrence wrote:
> 
> > Time per context switch = 688 nanosecs
> > 
> > [OK, this is a rather slow PC :-) ]
> 
> CommsTime in occam ...
> 
> Hmmm. This is what I got on a P5-133 with 128Mbyte DRAM:-
> 
> seq-ct:
> =======
> Time per context switch = 929 nanosecs
> 
> par-ct 
> ======
> Time per context switch = 1035 nanosecs
> 
> So this machine is even slower? Do these figures look reasonable?
> 
> Adrian
> -- 
> Dr A E Lawrence (from home)
> adrian.lawrence.without_spam@xxxxxxxxxxxxx
>
I get 

seq-ct
======
Time per context switch = 441 nanosecs

par-ct
======
Time per context switch = 518 nanosecs

on a P200MMX w/32MB EDO RAM (66MHz bus) - note _not_ a pentium 2 or
Pentium Pro - these will give better results, even at the same clock
speed.

These are comparable to Peter's first figure which was on the same
machine.  Better? 

Jim
---
Jim Moores, PhD Research Student, Computing Lab, University of Kent, UK.