Beowolf fun!

[Kevin D. Clark]

Kristian Erik Hermansen writes:

> You need to find a way to slice your
> computing application nicely so that the tasks can be broken down into
> very small pieces to be computed.

I'm sure that you know this, but, actually, in order to get reasonable
speedup, you want to be able to break up your problem in such a way
that (1) all of the processors are doing work (as much as possible)
(2) the units of work are as *large* as possible (3) you want to
minimize the amount of communications between each logical CPU
(communications kill speedup).  Satisfying all of these constraints is
always...interesting...

As for MPI implementations, I highly recommend MPICH.

Regards,

--kevin
-- 
GnuPG ID: B280F24E              God, I loved that Pontiac.
alumni.unh.edu!kdc                   -- Tom Waits

-- 
This message has been scanned for viruses and
dangerous content by MailScanner, and is
believed to be clean.