Beowulf usage statistics?
math at velocet.ca
Mon Jun 24 11:11:49 PDT 2002
On Mon, Jun 24, 2002 at 12:44:09PM +0200, Franz Marini's all...
> On Mon, 24 Jun 2002, Alan Scheinine wrote:
> > One possible explanation is that many small labs are planning
> > to build small clusters of PCs. If this is the case, the number of
> > clusters based on WWW documentation and based on talks at conference
> > may undercount the actual number by a factor of more than two or more.
> > (Analogous to saying that in terms of biomass and number of species
> > the dominant form of life on earth is bacteria.)
> I think you're right...
> Here at the Dept. of Physics of the University of Milan, we have at least
> 2 clusters, 1 16 diskless nodes +1 server, and 1 24 nodes + 1 server
> mantained by me :)
> So I think the real number of small clusters is actually much greater than
> the "official" one ...
Perhaps people have designed their clusters for actual use instead of
benchmarking to get into the rankings of the top N in the world, realising
their clusters wont perform well in that type of ranking.
Better to have no result than a low result confirming its slow for that
benchmark - fodder for the simpletons to use against you (especially
when budget monies are being squabbled over, it can get vicious).
Considering we design clusters for throughput instead of absolute speed
we'd never run any of ours vs that benchmark. Apples to oranges. Pointless.
Its just a testostermoronic size competition IMHO.
Ken Chase, math at velocet.ca * Velocet Communications Inc. * Toronto, CANADA
More information about the Beowulf