On Thu, 10 Jul 2014 17:51:36 +0000 Bob Tupper <bobctup...@gmail.com> wrote:
> > Greetings all. > > I have a requirement that I'm not sure how to fulfill and was hoping > to get some suggestions. > > each job requires > 8 cores on a single machine, 1 license and 128G of RAM (most > of my execution hosts are 16 cores 256G RAM) 4 or more simulations > jobs that must run at nearly the same time on any execution host. > > if the resources to run all 4 or more jobs are not available or will > not become available shortly, all of the jobs should pend until > resources become available. > > since these are very big jobs and must run at some point, how can I > make sure that the resources do become available at the same time > with out waisting resources? > > Any input would be very helpful and appreciated. > Thanks > -Bob I assume the runtimes of the jobs vary enough that submitting them as a single job to a PE with an allocation_rule of 8 would not work. Not wasting resources at all might be tricky but the following should produce minimal waste. Submit a short job (few seconds) that grabs the resources for all your jobs. Submit your real jobs with a dependency on said job and ensure they have high priority so they can grab the resources immediately after your lead job frees them. The high priority might need co-operation from the cluster administator if that's not you. William
signature.asc
Description: PGP signature
_______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
