[mpiwg-ft] FTWG Plenary Slides Part 1

[George Bosilca]

The proposed change to the agreement will allow the use of MPI_COMM_AGREE on a revoked communicator, and MPI_COMM_AGREE may return ERR_PROC if new failed processes are detected during the agreement. This allow the use of MPI_COMM_CREATE_GROUP followed by an MPI_COMM_AGREE on the original communicator. However, this solution might be suboptimal compared with the use of MPI_COMM_SHRINK, but is has the advantage of allowing the user to place the processes in the order it wants/needs without an extra MPI_COMM_SPLIT.

  George.

On Dec 4, 2013, at 21:32 , Wesley Bland <wbland at mcs.anl.gov> wrote:

> I don’t think that allows you to really use MPI_COMM_CREATE_GROUP. You still have the problem of an inconsistent communicator that you can’t use to validate or revoke the communicator. Did I miss something?
> 
> Wesley
> 
> On Dec 4, 2013, at 2:25 PM, George Bosilca <bosilca at icl.utk.edu> wrote:
> 
>> Just a side note. With the small change in the agreement we discussed lately the usage of MPI_COMM_CREATE_GROUP become an option. You might want to amend the slide 12.
>> 
>> George.
>> 
>> On Dec 4, 2013, at 18:58 , Wesley Bland <wbland at mcs.anl.gov> wrote:
>> 
>>> Hello WG,
>>> 
>>> As you know, we won’t be pursuing a reading at this meeting in an effort to better explain the proposal and finalize the improvements to the MPI_AGREE function. We will still be holding the plenary however and I’ve been working on the slides for that time. The current plan is to present the plenary in two parts. I’ll be presenting the first part which will include a modified version of the slides that were shown in Madrid. My understanding from talking to numerous people is that there were quite a few questions and misunderstandings at the last meeting and presenting those slides again with some updates would be valuable to address some concerns. I’ve attached the slides that I plan to present to this email (let me know if they get filtered). Everywhere that I’ve made an update or added a new slide, I made a note in the presenter’s notes so take a look and send back any notes you may have. We unfortunately won’t be able to have a con call about these slides before the meeting, but we will have time during the working group time on Monday that we can go over them if people would like to see them.
>>> 
>>> The second part of the plenary time will be presented by Aurelien. He’ll be giving examples of FT abstractions that could sit on top of the new ULFM mechanisms and how they could be implemented. I imagine more information about those slides will be coming later in the week as he finished them up.
>>> 
>>> Thanks,
>>> Wesley
>>> 
>>> 
>>> <2013-12-11 MPI Forum ULFM.pptx>_______________________________________________
>>> mpiwg-ft mailing list
>>> mpiwg-ft at lists.mpi-forum.org
>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-ft
>> 
>> _______________________________________________
>> mpiwg-ft mailing list
>> mpiwg-ft at lists.mpi-forum.org
>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-ft
> 
> _______________________________________________
> mpiwg-ft mailing list
> mpiwg-ft at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-ft