[PRL] Development of Large-Scale Parallel Scientific Codes

[Mitchell Wand]

Here's a very interesting article where the authors did case studies of 5
recent major parallel scientific codes.

The writing is terrible, but hold your nose and go through it to get a
picture of how things work in this area.

Take home messages (some surprising to me):

1.  everybody uses MPI.
2.  they don't optimize for particular processors.  The lifecycle for
processors is way too short.
3.  what they do have to do is work around compiler & system bugs on each
system.

Read the article, you are likely to find other nuggets. I've put up a copy
at
http://www.ccs.neu.edu/home/wand/private/Hochstein-Basili-IEEE-Computing-08.pdf

--Mitch
-------------- next part --------------
HTML attachment scrubbed and removed