Ticket #919 (new bug)
Opened 3 weeks ago
Allgatherv performance
| Reported by: | balaji | Owned by: | balaji |
|---|---|---|---|
| Priority: | major | Milestone: | mpich2-1.3 |
| Component: | mpich2 | Keywords: | |
| Cc: |
Description
One of the users reported a performance problem for Allgatherv for non-power-of-2 number of processes on large systems. On reading through the code, it looks like our algorithms rely on the "total message size" gathered, rather than the message size contributed by each process. So, for large-scale systems, even small message sizes go through an O(P) algorithm.
We should take another look at the irregular collectives to improve these cases.
Attachments
Note: See
TracTickets for help on using
tickets.
