about fast interconnects and SCI in particular

[Greg Lindahl]

> I would say that fine grain parallel problems are still best performed 
> on traditional supercomputers,

I would hesitate to jump to such a grand conclusion. Even for fine
grained programs such as Charmm, I've gotten supercomputer-beating
results with my AlphaLinux/Myrinet cluster.

> Finally, the more simultaneous users are allowed, the 
> less one should invest in the network, for the obvious reason that 
> the different concurrent applications are independent from each 
> others.  An expensive network is justified only if one must run 
> applications on all the nodes simultaneously. 

This, on the other hand, is a fairly good generalization, although I
wouldn't go so quite so far with it. For example, with fast ethernet
it's useful to run Charmm (molecular dynamics program) up to 4 cpus,
and with myrinet, up to 32 cpus. If you own 256 nodes and have 10
simultaneous users, Myrinet still wins. So, instead of your
generalization, I would suggest learning the characteristics of your
code, and use that knowledge to pick the right answer for your
situation.

-- greg