[Mpi-forum] MPI One-Sided Communication

[Greg Lindahl]

On Thu, Apr 23, 2009 at 01:28:22PM -0600, Simone Atzeni wrote:

> Hi, I'm an italian student and I'm doing a thesis about Formal  
> Verification. Now, I'm looking for information about mpi one-sided  
> communication to see who uses one-sided and for what (user applications 
> or implementing library).

As far as I can tell, usage of the MPI-2 one-sided library is about
zero. People tried it on various interconnects, and when it turned out
to be slower than 2-sided, stopped trying.

The only significant public one-sided app that I've run into is
NWCHEM, which uses a non-MPI one-sided library. And in that case,
PathScale did an implementation on top of 2-sided MPI-1 send/recv, and
it scaled better than competing real one-sided hardware. (Check out my
old whitepaper for numbers.)

I've gored enough oxen for one post, so I won't rain on your
model-checking parade ;-)

Now that I'm not an HPC guy anymore, I'll note that the distributed
database that I'm doing is completely implemeted in terms of active
messages.  They're easier to use corrrectly than real one-sided
communications, which are hard because programmers screw up
synchronization & double-buffering.

-- greg

p.s. I'm agreeing with Tony?!