On Dec 15, 2006, at 4:05 PM, George Bosilca wrote:
> Thanks for the graph. I was wondering how do we compare with the
> others MPI implementations. If you have others MPI installed on your
> cluster can you do the same measurements with them and augment your
> graph with the results ?
Here's a graph with Open MPI 1.2, MVAPICH2 0.9.8, and LAM/MPI 7.1.2.
It looks like derived datatypes don't have any real cost in this case
on any of these implementations. Open MPI 1.1 was *much* slower with
derived datatypes than with primitive/built-in data types, but I ran
out of space to show that here.
I hope that helps.