hw2 Program Exchange

[Matt Weiden (Reader)]

Hello Everyone,

Please send me one email with the names and emails of the people in
your group so that I can arrange the program exchange. After I have
paired up the groups, I will email you the contact info of the other
group you'll be working with.

Also: I've gotten quite a few questions about part 3 of the
experiments section (the tau question). I think that the interesting
things to look for here in your tau profile data is where the
bottlenecks emerge, as well as the proportion of communication time in
the overall runtime.

Looking at the Tau output you may find that some methods might take a
disproportionately large amount of the runtime. These would be the
methods to try to optimize first if you were tuning your code.

For a given problem size, if you vary the number of processors and use
tau to break down the overall runtime into computation time and
communication time you might notice an interesting trend like this:

http://cs.ucsb.edu/~weiden/cs240s2011/plot.jpg
*note this plot is hypothetical

There may be an optimal number of processes to break a problem into
given a certain problem size and the way the communication overhead
adds up.

Matt