[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Interfacing GALib to Swarm
From: |
John W. Fondon III (Trey) |
Subject: |
Re: Interfacing GALib to Swarm |
Date: |
Fri, 10 Oct 1997 16:43:55 -0600 |
>>
>> From: "John W. Fondon III (Trey" <address@hidden>
>> Subject: Re: Interfacing GALib to Swarm
>
> [stuff deleted...]
>
>> BTW, this reminds me of a question I have that is not (exactly) related:
>> What is the minimal system upon which SWARM can be executed?
>> I'm planning on stacking up cheap pentium motherboards, sharing a single
>> disk, monitor, etc.
>> Is this feasible?
>>
>> Trey
>
>A friend of mine in graduate school had a similar setup worked out, with some
>important differences: he was not using swarm, and he was stacking 386
>and 486
>boards (really cheap) rather than pentiums. He was using the setup for
>genetic
>algorithms for optimization problems in physics, and wished to have several
>processors working on the problem in parallel.
>
>Each board simply had a cpu, 4 or 8 M RAM, and a network card. Each board
>booted through its network card, started the appropriate daemons, and
>waited for
>instructions from the server (the sole pentium) to work on portions of the
>problem. All communication was done via TCP/IP between boards. The minimum
>requirements for each board were those required for a minimal linux kernel
>configuration, a few daemons, and preferably no swap space.
>
>I have never bothered to find out what the minimum configuration would be to
>have such a board running a linux kernel and a swarm process. Is it much more
>than is required for just a linux kernel?
>
Dr. Box,
Not only SWARM and the kernel, but also the code which will
evaluate fitness (of course, this can always be shipped over to the server
if that is more efficient).
I fairly certain this is the most cost effective way to tackle a hard
problem that lends itself to distributed computing...
I've priced out a 2 generation removed pentium/mb/netcard and it's
amazingly cheap when you compare cycles/second to alpha systems - a biggie
that I have not resolved yet is RAM/cache (this will depend on SWARM and my
problem). But I really think SWARM is the way to go as long as it will not
have any fundamental conflicts with this type of computing scheme...
==================================
Swarm-Modelling is for discussion of Simulation and Modelling techniques
esp. using Swarm. For list administration needs (esp. [un]subscribing),
please send a message to <address@hidden> with "help" in the
body of the message.
==================================