[Mpi3-hybridpm] Telecon 10/6 at 12pm central

James Dinan dinan at mcs.anl.gov
Thu Oct 6 14:03:11 CDT 2011


As I've written it, the result would be an implementation-defined 
default neighborhood.  I think it's challenging to get more specific 
than this.

We could include an advice to implementers to the effect that this 
split_type should return a communicator in which processes can 
communicate efficiently with one another.

  ~Jim.

On 10/6/11 1:09 PM, Pavan Balaji wrote:
>
> What's the behavior when MPI_INFO_NULL is passed?
>
> -- Pavan
>
> On 10/06/2011 12:38 PM, James Dinan wrote:
>> Hi All,
>>
>> We discussed adding another split_type to the MPI_Comm_split_type
>> proposal (ticket 287) to broaden the use of this routine from shared
>> memory to being able to split based on topology. Text would look
>> something like:
>>
>> MPI_COMM_TYPE_NEIGHBORHOOD - this type splits the communicator into
>> subcommunicators, each of which encapsulates a neighborhood of
>> processes. Implementations may provide info keys that can be used to
>> specify the criteria used to group processes, e.g. using a feature of
>> the system topology.
>>
>> Overall, we agreed that this would be a good addition. We want to
>> solicit feedback from other members of the working group (especially
>> Ron, who is the ticket's owner) as to whether we should add this to the
>> existing text or create a new ticket and pursue this independently.
>> We're planning another formal reading of this ticket at the October
>> meeting.
>>
>> Thanks,
>> Jim.
>>
>> On 10/6/11 10:28 AM, James Dinan wrote:
>>> Hi All,
>>>
>>> Quick reminder: we're connecting today at 12pm CT.
>>>
>>> ~Jim.
>>>
>>> On 10/5/11 2:57 PM, James Dinan wrote:
>>>> Hi All,
>>>>
>>>> It looks like tomorrow, October 6, at 12pm CT will be the best time to
>>>> get everyone. The telecon information is available on the wiki.
>>>>
>>>> https://svn.mpi-forum.org/trac/mpi-forum-web/wiki/MPI3Hybrid
>>>>
>>>> The agenda for this call will include:
>>>>
>>>> 1. Helper team proposal: We will discuss the result of the straw
>>>> vote on
>>>> whether we want to take the proposal to the Forum.
>>>>
>>>> 2. Update from Ron on the Comm_split_type proposal.
>>>>
>>>> 3. Update from Torsten/Jim on the shared memory windows proposal.
>>>>
>>>> 4. Discussion on the endpoints proposal.
>>>>
>>>> Talk to you then,
>>>> Jim.
>>>>
>>>> On 10/4/11 3:50 PM, James Dinan wrote:
>>>>> Hi All,
>>>>>
>>>>> I set up a doodle poll for this Thursday, Friday, and the following
>>>>> Monday. Please enter your availability:
>>>>>
>>>>> http://www.doodle.com/aah8tm3rbcgz7dmv
>>>>>
>>>>> Thanks,
>>>>> Jim.
>>>>>
>>>>> On 10/4/11 10:17 AM, Pavan Balaji wrote:
>>>>>> Hi Steve,
>>>>>>
>>>>>> On 10/04/2011 10:14 AM, Steve Oyanagi wrote:
>>>>>>> If I recall correctly at the last telecon on 9/28 we pushed several
>>>>>>> items off until the next meeting, which I thought someone said was
>>>>>>> next week (i.e. this week). Is there a meeting this week? Is there
>>>>>>> a regular telecon time these days?
>>>>>>
>>>>>> Sorry, I didn't get a chance to schedule one yet. I'll do that.
>>>>>>
>>>>>> We moved away from a regular telecon time as many people weren't
>>>>>> able to
>>>>>> make it at a fixed time. But we can go back to that if that's
>>>>>> preferred.
>>>>>>
>>>>>> -- Pavan
>>>>>>
>>>>> _______________________________________________
>>>>> Mpi3-hybridpm mailing list
>>>>> Mpi3-hybridpm at lists.mpi-forum.org
>>>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm
>>>> _______________________________________________
>>>> Mpi3-hybridpm mailing list
>>>> Mpi3-hybridpm at lists.mpi-forum.org
>>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm
>>> _______________________________________________
>>> Mpi3-hybridpm mailing list
>>> Mpi3-hybridpm at lists.mpi-forum.org
>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm
>> _______________________________________________
>> Mpi3-hybridpm mailing list
>> Mpi3-hybridpm at lists.mpi-forum.org
>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm
>



More information about the mpiwg-hybridpm mailing list