[RC5] Slow clients
jcampbel at lynn.ci-n.com
Mon Jul 13 11:06:39 EDT 1998
On Sat, 11 Jul 1998, Remi Guyomarch wrote:
> ? 6kkeys/s on DES ? I don't see how this could happen. Even if for some
> reasons the client selects the PPro/PII DES code (the worst of the twos
> on 486), you will get ~50 kkeys/s. Are you absolutely *sure* that
> nothing else is running in the background ? Check with "/usr/bin/time
> ./rc5des -benchmark" and look at the %cpu you get.
Well, there's always *something* running in the background... this
is a Linux server. I'm not sure, though, what it could have been that could
have bogged down the DES benchmark but not the RC5 benchmark twice in a row.
I ran the benchmark a couple more times yesterday and it came up with
100+kk/s for DES both times, which is more in line with what I expected. I'm
running the new client in the real world now, and the RC5 performance, at
least, appears to be normal. We'll see what we get when it switches over to
DES this afternoon.
> Same thing here, there is *no* code in this client that will runs at
> 6kkeys/s on RC5 with a 486 DX-33... '/usr/bin/time' the benchmark. Also
> post your "rc5des.ini" file.
> The worst thing you can select for a 486 is the K5 RC5 code. Expect ~26
> kkeys/s with this code on a 486 DX-33.
I'm still having some trouble with this one... it's running
10-15kk/s RC5 in real life with either the new client or the old one now.
The benchmark shows similar performance, with time claiming that it has
96-98% of the CPU. This is less than half the speed I'd expected. From my
logs it seems that the speed dropped less than a month ago from the 30kk/s
it had been getting down to around 8kk/s. I don't think I've significantly
changed the machine's configuration since then, and the performance drop
doesn't seem to have affected normal operations, like compiling and such.
The .ini is configured properly; it's been using pretty much the same .ini
since RC5-56. I ran through the options after installing the new client to
make sure nothing had changed, and it seems to be fine. And, yes, the turbo
switch is on. I checked.
> > The new client is an improvement on my AMD 386-DX-40. RC5 speed has
> > remained very much the same, but the client has quit dumping core every time
> > it tries to do DES. It remains to be seen if it'll hold up during the real
> > contest, but, for the first time, it made it through the DES benchmark
> > without crashing. The speed was reasonable, too.
> Which client version crashed on DES on your processor ?
I'm not sure of exact versions; it was several of the early 2700
series linux (both -mt and -non-mt) clients, from during DES-II-1. I finally
gave up, and that box continued to crack RC5 for the duration of the DES
contest. After DES was over, I didn't bother trying to get any of the newer
ones to work, figuring that an RC5-only client was good enough for doing RC5
jcampbel at lynn.ci-n.com
QotD: Sharks are as tough as those football fans who take their shirts off
during games in Chicago in January, only more intelligent.
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest
More information about the rc5