-
Notifications
You must be signed in to change notification settings - Fork 1
2020 04 20 webex
PR is at https://github.com/mpi-forum/mpi-standard/pull/160
This PR is ready to merge into master. There is a link to the delta between what was voted on at the Portland Forum (2/20) and what's at the HEAD of the PR in one of the PR comments.
Martin doesn't think we need to do another vforum reading in a prep for the "f2f" in June. Send message to the forum about the state of the PR.
Howard bring's up MartinR's comment.
Martin carries discussion forward. How to add more dynamic capability to the sessions model. Martin is okay with putting his slides on the WG wiki, but will do some cleanup first.
Discussion of bubbles as protection domains. If "bubbles" can be merged or expanded, that seems to muddy the idea of protection domains. Slide 5 in the current slide deck. Dan describes a "lock this set of processes" into a bubble concept. Martin gives example of a misbehaving in-situ viz tool. Dan thinks this isolation concept should fit here, but would need a "conduit" to connect to the actual simulation processes. See "reasons for locking of bubbles" slide in the slide deck. Scope of "seeing" process sets, for example across different systems.
Notion of implicit/explicit bubble creation discussed. Can we avoid explicit? How can we avoid explicit if we need a "lock down" notion. Martin adds a graphic to the slide deck to illustrate better the implicit bubbles model using comp and viz. Two bubbles - each "bubble" mapping to a session. Example of various users being able to attach/detach to the job dynamically. Example of ECMWF IFS app (https://www.ecmwf.int/en/research/modelling-and-prediction). This app uses realtime field data to "steer" the simulation.
Tony brings up the IMPI - apparently many variants what "I" means. Tony will put some refs on the wiki. Here is the remaining (online) information from NIST from 2000; there is also a paper, evidently: https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=151467 ; the standards doc is attached after the paper.
- Howard will do some gitology on the missing sentence in dynamic chapter that Martin had mentioned.
- Martin will upload slides to the wiki
- Tony will add some links to IMPI to the wiki.