July262016 - openpmix/openpmix GitHub Wiki

##PMIx Developers Meeting

July 26th, 2016

Ralph, Annu, David, Artem, Chris, Charles, Boris

• Question on query to get all nodes. This is not implemented today in PMIx. Should this be implemented in PMIx server library or RM. The RM already provides this info so they can query on PMIX_NODE_MAP key the library can return the regex provided by the RM.
• get optimization - requiring wild card rank to be specified for non rank specific stuff could be hard on the user. As they now have to think if a piece of data is rank specific or job specific. The community to expect the user to know the difference. Will keep the same for now until a use case comes where differentiating gets harder.
• Dave has a couple of fixes on get optimization, he will pr that to Ralph’s get optimization branch.
• Dynamic plugins in progress working on version1 - version 2 communication. Estimated time of completion end of the week.
• Optimization for scalable specifying large array of processes to PMIx. Proposal to change pmix_rank_t to uint32_t from int32_t and reserve higher values for groups like pmix_rank_local, pmix_rank_local_switch. We need to discuss further about use cases and may be provide an api to create a group if needed.
• POC for dealing with multiple versions of PMIx server plugin in RM seems to be a feasible solution. Dave and Artem to test on their RM packages as a next step.
• PMIx 2.0 is on track for release in early September. Code freeze in Mid August. Dynamic plugin and shared memory are the required features and the rest are nice to haves