[Beowulf] Cluster newbie, power recommendations

Robert G. Brown rgb at phy.duke.edu
Tue Mar 21 03:58:05 PST 2006


On Mon, 20 Mar 2006, Charlie Peck wrote:

> I think clusters like the one Eric wants to build have /significant/ 
> educational value, both in the building and the use.  How else does one learn 
> to do parallel/distributed programming if not on a cluster, even a "toy" one? 
> Sure the single AMD64 will be more powerful but it won't provide an 
> opportunity to learn about message passing, speedup, efficiency, problem 
> decomposition, etc.

Oh I agree, but I do think it is fair to point out the $1/watt/year
problem.  That's enough to at least open up the possibility of e.g. a
four node AMD64 cluster with modern CPUs, memory, NICs instead with the
money you save.  It's a trade-off -- a smaller, but much faster cluster
doesn't demonstrate scaling over as long a range and so on, but it does
MUCH more actual computation per second if there is any chance that he's
building a cluster to eventually try to do some real work.  But then,
clustering is always about trade-offs;-)

> There are a couple of issues when ganging multiple power supplies, most 
> having to do with the lines in the ATX standard where the board sends a 
> signal back to the power supply.  I believe the site that RGB is referring to 
> is http://joule.bu.edu/~hazen/LinuxCluster.  If this isn't enough get in 
> touch, we have worked through these same problems for our portable clusters, 
> see http://cs.earlham.edu/little-fe

Yeah, that looks like it.  Kinda cute in its own way.  And you don't
"have" to buy stock aluminum or steel and cut and bend it into shelf
trays for the motherboards, BTW -- there was a design somewhere that
mounted the motherboards onto cookie sheets and "racked" them in a
standard cookie-sheet rack, or a rack like the ones they use in
cafeterias.  Slide in, slide out.

    rgb

>
> charlie
>

-- 
Robert G. Brown	                       http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at phy.duke.edu





More information about the Beowulf mailing list