[Beowulf] multi-threading vs. MPI

[Greg Lindahl]

On Tue, Dec 11, 2007 at 08:52:11AM -0500, Joe Landman wrote:

> On the contrary, it is precisely because people are asking "how should I 
> parallelize" that they need to ask the basic question of "where does my 
> code spend time for my problems."

OK, so say I have a garden-variety finite-difference code. I know how
to use OpenMP to parallelize all the loops, and how to use halo
exchange with MPI to parallelize it. And maybe I'm even clever enough
to know how to use halo exchange with OpenMP, which is pretty ugly
code, but has better locality and scalability.

Before I've coded all 3 up, how does that help me pick which method to
use? It doesn't. And that's the question people are asking: should I
spend the time to implement X or Y in the hopes it'll be faster than
my existing code Z?

-- greg