[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: execution time
From: |
Jan Kreft |
Subject: |
Re: execution time |
Date: |
Wed, 6 Dec 2000 08:56:12 +0000 (GMT) |
Hi Steve,
On Tue, 5 Dec 2000, M Lang / S Railsback wrote:
> "Marcus G. Daniels" wrote:
> >
> > >>>>> "PT" == Pietro Terna <address@hidden> writes:
> >
> > PT> Is it useful to invest some time experimenting the Beowulf
> > PT> project to run Swarm on a cluster of Linux PC?
> >
> > I would instead suggest investigating Mosix (http://www.mosix.org).
> > It's a kernel extension to Linux that makes process migration
> > transparent.
>
> Let me ask a related question- what might be a good hardware solution to
> the problem of models with lots of agents doing lots of things, assuming
> no parallelization? A year or so ago somebody recommended DEC Alpha
> workstations as a good workhorse; meanwhile there are faster Intel
> machines out. Does anybody have any sound recommendations?
>
> Is CPU speed the only real issue, assuming you can always add enough
> RAM?
>
> Under what conditions is it a big advantage to have two processors? I
> suppose multiple model runs can be shipped to different processors- does
> a workstation running NT do stuff like that automatically?
Absolutely. I'm glad I've got 2 cpus, this effectively halves the time
needed before a bunch of runs is finished. And I suspect that you, like
me, almost always make many runs that will be compared later. So, with
several cpus in one box, you don't have to invest any time in ||lization
or getting a cluster organized, yet, with a bunch of runs, the time is
divvied up nicely. And one cpu more doesn't make the machine much more
expensive. I really recommend that. As a side effect, the system is pretty
responsive even under high load, say 4 simulations simultaneously, if you
niced them. Don't forget that you need more memory for simultaneous runs,
but thats probably obvious.
Best, Jan.
==================================
Swarm-Support is for discussion of the technical details of the day
to day usage of Swarm. For list administration needs (esp.
[un]subscribing), please send a message to <address@hidden>
with "help" in the body of the message.