On Wed, 12 Dec 2001, Brian Chase wrote:

> It also becomes pretty trivial to manage and scale a very large cluster
> of loosely coupled Linux or *BSD PCs.  Assuming they've the same
> software config across the systems, you can just net install the
> critters.

But the problems that start occuring with clusters of that size involve
inter-machine communication.  That's why mainframes are so efficient with
I/O.  The communication between CPU's (indeed, communication internal to
CPU's) and between CPU's and I/O controllers is very well engineered.  If
you want processor cycles, go buy a supercomputer.  I wonder how the TCO
for a used SP and a used Mainframe would stack up against 6000 PC's.

And, there are *lots* and *lots* of mainframe programmers out of work
around here who don't mind putting hours in. 8-)

Peace...  Sridhar