Is there a way to override that behavior? On occasion we have nodes go down, and since we're more development then production our users routines salloc nodes and then ssh to that node. I'd prefer they start from 1 rather then 86 (in this instance)
we're running a flat topology, core/edge on ethernet and infiniband (but have not configured any topology/tree stuff in slurm) On Fri, Feb 18, 2011 at 9:12 PM, Jette, Moe <[email protected]> wrote: > SLURM uses a best-fit algorhthm. > If you have topology/tree configured, the best-fit applies on a per-switch > basis. > Otherwise it applies to the 1-D list of nodes. In your case, you want 1 node > and > the choice is between a block of 84 nodes (1-84) or 15 nodes (86-100), so it > picked from the block of 15 nodes. > ________________________________________ > From: [email protected] [[email protected]] On > Behalf Of Michael Di Domenico [[email protected]] > Sent: Friday, February 18, 2011 12:35 PM > To: slurm-dev > Subject: [slurm-dev] node allocation choice > > How does slurm determine which node to use from a given node pool? To > give a more specific example, if you had nodes 1 through 100, but node > 85 was "down*". What causes slurm to grab node86 next instead of > node1? > >
