Hi, working on simulating large clusters, I experienced a slow down in the creation of the topology information in slurmd that is due to the fact that the node hash table was not build prior to that call. You will find enclosed a very small patch that add that. If I remember well I was at the origin of the addition of the build of the topo data in slurmd (to push the topology information in the job env) so I can only blame myself :)
Regards, Matthieu
From 350274d422d31979abb105b6b0e4bb113598a3ab Mon Sep 17 00:00:00 2001 From: Matthieu Hautreux <[email protected]> Date: Wed, 25 Jan 2012 22:28:51 +0100 Subject: [PATCH 1/1] slurmd: build node hash table before the topology data to speed up the process --- src/slurmd/slurmd/slurmd.c | 2 ++ 1 files changed, 2 insertions(+), 0 deletions(-) diff --git a/src/slurmd/slurmd/slurmd.c b/src/slurmd/slurmd/slurmd.c index a17311d..298381c 100644 --- a/src/slurmd/slurmd/slurmd.c +++ b/src/slurmd/slurmd/slurmd.c @@ -1228,7 +1228,9 @@ _slurmd_init(void) /* * Get and set slurmd topology information + * Build node hash table first to speed up the topo build */ + rehash_node(); slurm_topo_build_config(); _set_topo_info(); -- 1.7.6.2
