large clusters and topologies

Joel Jaeggli joelja at
Fri Jul 28 10:49:01 PDT 2000

On Fri, 28 Jul 2000, Greg Lindahl wrote:

> > - I suppose Ethernet is not all too efficient with this
> > size, even with switches instead of hubs. What's the largest
> > practical size for an Ethernet 'wulf?
> Depending on the details, Ethernet can work great for larger machines. The
> 1,000 cpu genetic algorithm machine ( uses
> ethernet, Google's 5,000 cpu cluster uses ethernet, Incyte's 3,000 cpu
> cluster uses ethernet, and so forth. The key is that their application's
> network bandwidth and latency demands aren't maxing out what they have.

that and you can now get fast-ether switches with port-counts up to about
384 ports which isn't to shabby compared to the denisty of most other

as greg describes the real issue of it this is worthwhile or not is
largely a question of where your bottlneck is...
> > Any alternatives to Myrinet upwards of that?
> Not really. It depends.
> > - Which type of algorithm can take direct advantage of a cube,
> > toroid, ... topology? I would think even a "dumb" job like
> > matrix multiplication might benefit from a 2D lattice (instead
> > of the typical all-to-all unique segment). Comments?
> There are lots of books which discuss this topic. There have been many
> religious wars fought over the subject.
> -- greg
> _______________________________________________
> Beowulf mailing list
> Beowulf at

Joel Jaeggli				       joelja at    
Academic User Services			     consult at
     PGP Key Fingerprint: 1DE9 8FCA 51FB 4195 B42A 9C32 A30D 121E
It is clear that the arm of criticism cannot replace the criticism of
arms.  Karl Marx -- Introduction to the critique of Hegel's Philosophy of
the right, 1843.

More information about the Beowulf mailing list