Hello, thank you to both people who replied in the meantime.
On Thu, Sep 01, 2016 at 07:44:30AM -0700, Robbert Eggermont wrote: > > On 01-09-16 13:06, Christof Koehler wrote: > >"The only changes, users see when using the select/cons_res plugin, are > >that jobs can be co-scheduled on nodes when resources permit it." > >which taking [2] and manpage of sched.conf (OverSubscribe defaults to > >"NO") into account means that in your configuration only jobs from > >different partitions/queues can share a node. > > Incorrect: jobs from one partition can share a node. Well, of course: A "node" is undefined in the select/cons_res context, it is only defined in the select/linear context, at least in http://slurm.schedmd.com/cons_res_share.html. I was not very exact. > > >Now assume that there is only a single partition, which I think we would > >prefer ad hoc. Then: > > > >1. The default (select/linear, OverSubscribe=NO) is always exclusive. So the > >job of course gets billed for the whole resources of a node. > > Correct. > > >2. With a single partition select/cons_res and OverSubscribe=YES must be > >simultaneously set so that jobs can share nodes. A user may allow other jobs > >on a node allocated to his job by explicitely setting "--shared", in which > >case > >only the actually allocated resources are billed to each different job on > >that > >node. If he does not share the full resources are billed to him. The option > >"--exclusive" is never needed. > > Incorrect. OK this is probably partly due to the same inaccuracy in my terminology as above. > > It's like this: > "select/cons_res" enables splitting of the resources (CPUs) between > jobs (so multiple jobs run on a node, each with dedicated > resources). Only the dedicated resources are billed. With > "--exclusive" the whole node is dedicated (and billed) to one job. > > "OverSubscribe=FORCE" (or "OverSubscribe=YES" with "--shared") > enables sharing of the resources (CPUs) between jobs (multiple jobs > run on the same CPU, using time-sharing). A job is billed for the > time it used the dedicated resources. With "--exclusive" the whole > node is dedicated (and billed) to one job. Yes, it is simple if someone tells you the right answer :-) I was not thinking clearly about the meaning of the word oversubscribe and the got completely confues by mixing up nodes versus cores/cpus/sockets. > > >4. Only if OverSubscribe=FORCE is set on the partition the argument > >"--exclusive" would make sense to prevent the default sharing of nodes. > >With "--exclusive" all resources of the node would be billed to the > >exclusive job automatically, right ? > > Correct. Thank you both very much for your answers and hints. So this behaves exactly as most people would like it once you understand it. Best Regards Christof > > Best, > > Robbert > > -- > Robbert Eggermont Intelligent Systems > [email protected] Electr.Eng., Mathematics & Comp.Science > +31 15 27 83234 Delft University of Technology -- Dr. rer. nat. Christof Köhler email: [email protected] Universitaet Bremen/ BCCMS phone: +49-(0)421-218-62334 Am Fallturm 1/ TAB/ Raum 3.12 fax: +49-(0)421-218-62770 28359 Bremen PGP: http://www.bccms.uni-bremen.de/cms/people/c_koehler/
signature.asc
Description: Digital signature
