SMP robustness

[Greg Lindahl]

> 	Beowulf systems, viewed as a lot of little independent
> hardware pieces,
> are quite more solid from that point of view. Like a bunch of ants.

Yes, this is our experience. The FSL cluster is extremely rarely down. The
number of pieces of hardware in single points of failure is pretty small.
And we are making that number smaller over time, while SGI is making their
number bigger with bigger machines.

-- g