Project

General

Profile

Task #2376

max_mpi_ranks has wrong name

Added by Mark Abraham 9 months ago. Updated 8 months ago.

Status:
New
Priority:
Normal
Assignee:
-
Category:
testing
Target version:
-
Difficulty:
uncategorized
Close

Description

regressiontest complex/orientation-restraints sets max_mpi_ranks to 1, because it cannot support domain decomposition, but it can and should be able to run with a separate PME rank, which the control logic does not account for. This exposed an issue for https://gerrit.gromacs.org/c/7447/ but it is not yet clear where the real problem lies.

Anyway, the max_mpi_ranks should instead refer to domain decomposition, since that is the intended use

History

#1 Updated by Berk Hess 9 months ago

Note that there is also some confusion around "domain decomposition". There is also domain decomposition for PME and the PP and PME decompositions are closely linked. I don't have a good suggestion for clear naming for the whole domain decomposition and the PP and PME domain decomposition considered separately.

#2 Updated by Szilárd Páll 9 months ago

  • Description updated (diff)

#3 Updated by Szilárd Páll 9 months ago

Mark Abraham wrote:

This exposed an issue for https://gerrit.gromacs.org/c/7447/ but it is not yet clear where the real problem lies.

PS to make edit clear: it turned out that the issue was in fact related to the fix and not to the rest of the code.

Note that there is also some confusion around "domain decomposition".

Indeed, I ran into the same issue thinking that in the GROMACS lingo "domain decomposition" refers to exclusively to PP decomposition (and that the "Initializing domain decomp" message in the log file was perhaps incorrect).

Also available in: Atom PDF