Beowulf flops info.

[Greg Lindahl]

> I just wondered if anybody has made a study on cost/performance ratio
> between beowulfs and commercial supercomputer?.
> any pointers?

I have never seen a GOOD comparison. There are 2 possible flaws:

1) Forgetting to include the labor needed to build the beowulf. Sometimes
that labor is free (I was once a grad student), but it should still be
included.

2) Comparing apples and oranges, especially with respect to network. Many
supercomputers have a pretty good network, while most beowulfs have fast
ethernet. If your problem doesn't need the network, the comparison makes
sense, but if you're trying to compare your beowulf to a supercomputer
running problems that NEED that faster network, the comparison doesn't make
sense.

If you get these right, then what's the FSL system counted as? It's a
commercial Linux Alpha cluster... the problems running on it need a fast
network, and there was no free labor.

-- g