[rc5] Java and RC5

Marc Sissom msissom at dnaent.com
Tue Nov 11 10:36:50 EST 1997

This mailing list is weird. Here I see a reply to my post, when
my own post has not yet come back!

Joseph Fisk wrote:
> On Tue, 11 Nov 1997, Marc Sissom wrote:
> >         136886 keys/sec
> That's the best result I've seen for any machine so far for the Java
> "client".
> > These results from a ppro 233, nt4, nsc4.01, running as a light
> > weight disk server, pproxy server for rc5, internal web service,
> > etc. 'Normal' perf for the rc5 w32gui is ~500K.
> PPro 233?  Did you clock it from 200mhz or from 180?

>From 200. No problems at all. No heat, no errors, on an asus
p6np5 mb just move a jumper and turn the power back on.
> A PPro 200 scores over 510kkeys in Linux; a 233 should score some
> 590kkeys.

I have been running the service client for nt4, but I do know that
the os overhead on w32 machines is greater than for linux. In my
experience, linux and os2 are neck and neck, FreeBSD is the fastest,
and w32 is the slowest, with NT having an advantage over w95 on 32
bit optimized machines like the PII or PPro, and w95 having the
advantage on 16 bit machines.

The client can do better. On a pp200 with NT4 the score of the
w32gui is ~490K. This pp233 runs ~15% faster, but has more
extraneous load so...it seemed that 4:1 ratio was holding.

> I have seen a benchmark for a PPro240 as well but it seemed low for
> the Mhz rating.

In my experience with varying proc clock rates, bus clocks,
wait states, etc. you _never_ want to decrease the bus speed.
I assume that it was 60x4. It takes a big boost in processor
speed to offset the lost time on bus cycles. 66x3.5 will almost
always beat that 60x4 arrangement. Apparently even in a proc
intensive benchmark like the rc5.

To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.

More information about the rc5 mailing list