large clusters and topologies
award at mypic.ad
Fri Jul 28 05:33:31 PDT 2000
This discussion we've been having about > 100 node clusters
makes me wonder about two points:
- 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? Any alternatives to
Myrinet upwards of that?
- Which type of algorithm can take direct advantage of a
toroid, ... topology? I would think even a "dumb" job like
matrix multiplication might benefit from a 2D lattice
of the typical all-to-all unique segment). Comments?
More information about the Beowulf