[Mpi-forum] Results from the Albuquerque MPI Forum (INCLUDES ACTION ITEM)
schulzm at in.tum.de
schulzm at in.tum.de
Sat Dec 14 14:54:18 CST 2019
(Warning: long email with action items at the end)
We had a (at least in my eyes) very successful meeting last week - thank you to all who were active in the many discussions and/or helped with preparations. In the following a quick summary:
Several tickets passed the reading stage, including MPI Sessions, Partitioned Communication, and additions for hardware topologies. Those will come up for 1st votes next time in Portland, matching our timeline that we had set for MPI 4.0. Additional votes and NoNo updates for tickets are listed on the voting page for the meeting.
We also had discussions on changes on how the bindings in the document are built (aka. Pythonization) and on many smallish clean-up changes that Bill accumulated. Both will require review by the chapter committees and we have discussed a process for this review - Jeff will send around explicit instructions for chapter committees this week - please take a close look and try to stick with the deadlines in that email - we need to get this done in time for the Portland meeting (to be precise 4 weeks BEFORE the Portland meeting).
To ensure that we have active committees as well as chairs who are familiar with the process as discussed, we shuffled the chapter committees around - please take a look at this on the MPI 4 page - if you would like volunteer for an additional chapter or if you feel your name shouldn’t be there, please contact the chapter committee chair (with Wesley and myself on cc).
Overall, we decided to stick with our timeline (of course, pending review in Portland), as follows:
Feb 2020: Final 1st Votes
June 2020: Final 2nd Votes
Sep 2020: Ratification Meeting
Nov 2020 (special meeting at SC): Final ratification
This timeline, however, leaves two issues out, which weren’t quite ready in Albuquerque:
- BigCount - we have more or less reached consensus on the text changes, but the new bindings are missing
- Terms and Semantics - long discussions and still many changes pending
With the above timeline and the changes compared to the last officially announced versions likely not passing a NoNo vote, these two issues would not make MPI 4.0. As some members felt that those two items need to go out quickly as well, but overarching opinion was that we made a firm commitment to the community for an MPI 4.0 at SC2020, we discussed two alternatives:
A) Immediately follow up with a clean-up 4.1 version with final 1st votes in September and final 2nd votes in November, so that we can have a draft 4.1 along the new MPI 4.0 with BigCount and clean-ed up terminology (plus several clean-up tickets stemming from Bill’s full document review) as well as possible some restructuring, but no new functionality beyond BigCount.
B) Add another meeting in the April and/or August timeframe to buy us one more meeting for discussions.
ACTION ITEM BELOW:
If there are any strong opinions one way or the other, please let me know (I would suggest to send this directly to me and not over the whole list to give everyone equal chance to speak up and I will collect all input). Also, as both tickets are controversial, I would like to encourage all active members of the forum to a) actively participate in the discussion so we can have productive discussions at the meetings and b) have internal discussions in the respective organizations to form an opinion on where they stand on the two tickets (both in general and how critical they are to the MPI 4/4.1 timeline) - also here, please feedback to me and I will collect.
We will also have several virtual meetings on these topics (as well as the cleanup pass mentioned above) in the usual slot starting in the new year (we are skipping this week and - obviously - during the holidays). Please stay tuned for details and make every attempt to attend. It will be important to have good representation!
I am sure we will find a good solution to these issue and overall I think we are on track to a significant and strong update to MPI in 4.0 (and possibly 4.1). Thanks to everyone for their hard work on this!
Prof. Dr. Martin Schulz, Chair of Computer Architecture and Parallel Systems
Department of Informatics, TU-Munich, Boltzmannstraße 3, D-85748 Garching
Member of the Board of Directors at the Leibniz Supercomputing Centre (LRZ)
Email: schulzm at in.tum.de
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the mpi-forum