[geeks] distributedfolding

[Kurt Huhn]

Dave McGuire <mcguire at neurotica.com> wrote:
> > my Octane (note to self, download update), and a few Linux boxen.
> 
>    Cool.  How well does it run on your Octane?
> 

Amazingly well.  Depending on actual structure generated it can do about
1500 of the current protein model per hour.  Most of the time doesn't
seem to be spent actually generating structures, but in saving the
structure and prepping to build the next one.  There's a very long delay
there between structures.

This is measurably faster than my 'Blade 100 at work.  I forget the
actual number, but my R12K Octane stomps it.

>    I'm now running three instances of it on a 3-proc E450, and two 
> instances of it on a dual-proc Ultra60.  I will probably put it on my 
> I^2 R10K/195 and maybe on a dual-proc Ultra2 as well, though its only 
> got 167MHz processors.
> 
>    I wonder if they'd listen to requests for a NetBSD/alpha client 
> distribution.
> 

I wonder.  They might, if they get enough requests.  I've heard that the
Alphas do *amazingly* well at this.  Unforch, Tru64 is the only client
for Alpha, IIRC.

Heck, I'm running two on a Dual PPro-200 full time - every little bit
helps.  I'm also running it on my craptop (PII 266) and it's doggy, but
it's the only thing it really has to do, so...

I have it running on one processor of a dual proc P3 1.13Ghz server, and
it absolutely flies.  Load average is fucking up there, and BB
constantly complains about it, but it's crunching 'em down.  A good use
for x86 hardware!
-- 
Kurt                "I remember that I've got a solid base of worms and
kurt at k-huhn.com      ant eggs.  I puke like a hero all night long."
                                                      -- Tony Bourdain