At 01:04 PM 5/23/2006, Nathan Moore wrote:
Hi,
Just building a beowulf is in itself is a challenge - I'm impressed
that you're thinking about getting one together. As others have
said, Beowulfs are only useful if you have a tough problem to work on
(which also interests you).
There are a variety of "interesting" problems that are computationally
intensive. Some architectures might be optimized better than others, but,
in general, making anything make use of multiple processors in parallel is
a useful exercise.
1) Many game playing algorithms lend themselves to parallelism to explore
the move tree (Chess is iconic, as is Checkers)
2) Various optimal path routing algorithms can be parallelized (determining
the optimal route for an airplane, given a database of wind speeds vs
lat/lon/altitude). There's a well known optimal algorithm (with some
constraints) for the single processor case (the A and A-star algorithms),
but using multiple processors effectively is another matter.
3) All manner of pattern matching (genome, cryptography, etc.)
4) 3-d image rendering
5) Monte carlo simulations of many kinds
6) various and sundry differential equations on a grid solvers (Navier
Stokes for CFD, Maxwell's equations for electromagnetics, etc.)
Just getting your head around the idea of passing messages around to solve
a bigger problem is a useful thing to learn.
_______________________________________________
Beowulf mailing list, [email protected]
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
