If you specify —slot-list 0,2,4,… on the original mpirun cmd line, we will use only those cpus for both jobs. However, we will error out if more than one proc gets bound to a given core, so you’d have to also add —bind-to core:overload
> On Sep 22, 2015, at 12:40 PM, Glynn Hosking <glynnhosk...@yahoo.co.uk> wrote: > > I am using mpi_comm_spawn to launch one mpi-enabled executable from another. > The jobs are running on clusters using SGE and PBS schedulers. > > I can see that it's possible to specify the nodes on which to run the spawned > executable (using the info argument list). However, is it also possible to > specify the cores on which to run the spawned executable? > > I'd like the spawned executable to run on the same nodes and cores as the > master executable, to minimise the resources used on the cluster. The master > process would otherwise be doing "nothing" whilst waiting for the spawned > executable to complete its work. > Sent from Yahoo Mail on Android > <https://overview.mail.yahoo.com/mobile/?.src=Android>_______________________________________________ > users mailing list > us...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users > Link to this post: > http://www.open-mpi.org/community/lists/users/2015/09/27647.php
