[Mpi3-rma] RMA proposal
balaji at mcs.anl.gov
Sun Feb 28 22:14:41 CST 2010
I have received a few comments so far, mostly on additional text to
clarify a few things. Here's a new draft that adds these explanations
(no semantics changes compared to 1.5):
I urge everyone to please take a look at the document if you are
interested in participating in the RMA discussion during the next Forum
meeting. Trying to understand the semantics at the meeting would be a
waste of time for everyone.
On 02/24/2010 02:16 PM, Pavan Balaji wrote:
> Based on some additional internal discussion and comments here, there's
> a new version of the draft put up:
> -- Pavan
> On 02/23/2010 12:51 PM, Pavan Balaji wrote:
>> I've created a new RMA proposal
>> (http://www.mcs.anl.gov/~balaji/tmp/rma/rma-1.4.pdf) and would like to
>> initiate some discussion on it.
>> The general idea is similar to the lock/unlock primitives in MPI-2.2,
>> but the main difference is that it decouples synchronization of requests
>> from lock/unlock itself. I've added some discussion items at places
>> where there is no clear answer one way or the other. There are some
>> examples in the end to give you a feel of how it can be used.
>> In order to not confuse with the semantics of MPI-2.2, I've forked off a
>> new concept called "Portals" which is similar to "Windows". We can
>> either keep it as a separate chapter (or sub-chapter within RMA); or we
>> can decide to merge it back into the concept of "Windows" as a new
>> synchronization primitive. Names and syntax are always open to change.
>> Look forward to your comments.
>> -- Pavan
More information about the mpiwg-rma