[Mpi3-rma] MPI-3 UNIFIED model clarification
sayantan.sur at intel.com
Mon Jul 29 17:14:13 CDT 2013
> > I agree with the sentiment, also with the implementation issue of
> > UNIFIED with multi-rail. I'm not certain that a model exists in
> > between UNIFIED and SEPARATE. Either you are able to observe changes
> > in memory without further MPI calls, or not. Pavan do you have a
> > specific model in mind?
> Yes. I'd like a model where the user has to explicitly do a WIN_SYNC at the
> target. I agree that UNIFIED's current definition is stronger than this. Hence
> the request for either a new memory model or to weaken UNIFIED's
My vote is that we don't touch UNIFIED (as defined). If the UNIFIED model is useful, which I think it is, it will provide motivation for architectures/implementations that don't support it currently add special features for network synchronization.
If we require apps working with UNIFIED model to call MPI_Win_sync, an one-sided PGAS language might be forced to call sync for any buffer that may have been touched remotely. Thereby potentially causing some performance degradation.
More information about the mpiwg-rma