-
Notifications
You must be signed in to change notification settings - Fork 1
2025 05 19 webex
Continue with “dynamic processes with pset” proposal Attendance somewhat limited today.
Pick up on the MPI_Session_ipsetop discuss. Howard points out some of the questions left open last time. We start today with Dominik going over some examples he had added to the text since last time. Notion of the non-uniqueness of mpi://world is addressed in the example by an alias mechanism. Need a unique string/name for the dictionary.
Michael R. suggests MPI_Dict_setup would be better names MPI_Dict_create.
Discuss what happens if a process is not a member of the "mpi_main" process set. Would MPI_Sesssion_get_pset_info return something in that case?
A todo is to look at how the runtime implementations would be handling lookup of process sets.
Again defer persistent discussion till when Martin Schreiber can joint.
Go back to deconstruct of MPI_Session_ipsetop options. Sonja suggests a method for querying the runtime for suggestions about shrinking/growing. Discuss runtime vs application driven resource allocation changes. Could all this be handled by the col_info argument? Sonja would prefer a solution that doesn't require this.
Howard asked could we fire up the AI agent next time. Dominik will look in to this.
We will meet again on 2 June to continue discussions.