[Mpi3-rma] [EXTERNAL] Re: MPI-3.1 consideration slides

Barrett, Brian W bwbarre at sandia.gov
Tue Dec 4 11:36:05 CST 2012


Hubert -

I think one of the problems is that we're not experts at non-cache
coherent systems at this level.  Can you speak about issues you would see
in extending the shared memory windows to the separate model?  I'm not
against the extension; I think we largely didn't do it in MPI-3 because we
didn't feel we had the expertise to do it right.

Brian

On 12/4/12 2:19 AM, "Hubert Ritzdorf" <Hubert.Ritzdorf at emea.nec.com> wrote:

>Hi Jim,
>
>SX vector nodes are large shared memory systems. Therefore, I assume that
>users are quite interested in shared memory windows.
>
>Hubert
>________________________________________
>From: mpi3-rma-bounces at lists.mpi-forum.org
>[mpi3-rma-bounces at lists.mpi-forum.org] on behalf of Jim Dinan
>[dinan at mcs.anl.gov]
>Sent: Monday, December 03, 2012 5:44 PM
>To: mpi3-rma at lists.mpi-forum.org
>Subject: Re: [Mpi3-rma] MPI-3.1 consideration slides
>
>Hubert,
>
>Thanks for your comment -- in general, the RMA group is thinking about
>the vector systems.  Is there a desire from users to have shared memory
>windows defined for this platform?  Keeping in mind that the
>synchronization required will be harder to reason about than get/put?
>
>  ~Jim.
>
>On 12/3/12 10:24 AM, Hubert Ritzdorf wrote:
>> NEC SX (including the Earth Simulator) memory is not cache coherent and
>> MPI-2 RMA is working since many years on these systems.
>>
>> Hubert
>> ________________________________________
>> From: mpi3-rma-bounces at lists.mpi-forum.org
>>[mpi3-rma-bounces at lists.mpi-forum.org] on behalf of Jim Dinan
>>[dinan at mcs.anl.gov]
>> Sent: Monday, December 03, 2012 4:39 PM
>> To: mpi3-rma at lists.mpi-forum.org
>> Subject: Re: [Mpi3-rma] MPI-3.1 consideration slides
>>
>> On 12/2/12 7:44 AM, Jeff Hammond wrote:
>>>>> (1) Define semantics of shared memory windows in separate model.
>>>>>
>>>>> In Section 11.2.3 it says "MPI does not define semantics for
>>>>>accessing
>>>>> shared memory windows in the separate memory model."
>>>>>
>>>>> I would like to try to define these semantics in MPI 3.1.  Shouldn't
>>>>> it at least be possible to use MPI_WIN_(UN)LOCK with
>>>>> MPI_LOCK_EXCLUSIVE here?
>>>>>
>>>>> I believe that supporting RMA for weakly- or non- coherent memory
>>>>> architectures is vital for future systems.
>>>> I agree. We had this discussion and decided to postpone it because we
>>>> had other more pressing issues. Now may be the time. A discussion
>>>>would
>>>> be good.
>>>
>>> Yes, there was no urgency.  I'm glad the standard left this issue wide
>>> open instead of defining something restrictive.
>>
>> IMHO, defining shared memory windows in the separate model is not a
>> worthy use of the Forum's time.  This will be hard to get right, and
>> AFAIK no (reasonable) system requires it.  As we've currently left it,
>> if this functionality is needed, one can define the semantics as an MPI
>> extension.  I think this is a better option that defining a new model in
>> the absence of a platform that uses it.
>>
>>    ~Jim.
>> _______________________________________________
>> mpi3-rma mailing list
>> mpi3-rma at lists.mpi-forum.org
>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>>
>>
>>   Click 
>>https://www.mailcontrol.com/sr/BMhj+WAEk+nGX2PQPOmvUpFBc4ZD8M8UQDU1HKsp9B
>>M+kyw4sOr0J9g!Ah9tEzkBYIOMBNusRzyg1M3YVHc55g==  to report this email as
>>spam.
>>
>> _______________________________________________
>> mpi3-rma mailing list
>> mpi3-rma at lists.mpi-forum.org
>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>>
>_______________________________________________
>mpi3-rma mailing list
>mpi3-rma at lists.mpi-forum.org
>http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>
>_______________________________________________
>mpi3-rma mailing list
>mpi3-rma at lists.mpi-forum.org
>http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>
>


--
  Brian W. Barrett
  Scalable System Software Group
  Sandia National Laboratories








More information about the mpiwg-rma mailing list