You may want to look at the —oversubscribe mpirun option. If you want more control, you can consider making a rankfile where you place explicitly processes.
Aurélien > Le 29 juin 2016 à 11:50, Jason Maldonis <maldo...@wisc.edu> a écrit : > > Hi everyone, > > I am having trouble developing a complicated parallelization algorithm with > MPI and I'm hoping for some tips (I am using OpenMPI 1.10.2). I posted the > latest problem I ran into on Stack Overflow and got a response from someone > saying they don't think it is possible to do the spawn allocation that I want. > > You have all been very helpful in the past, so I am hoping to get your > opinions. > > Briefly, I am trying to oversubscribe cores to run spawned processes while > the main program's processes sit and wait in a barrier. Because the syntax > highlighting is better, it might be easiest to look at my post on Stack > Overflow than me trying to paste it in an email. Please see this link: > http://stackoverflow.com/questions/38087327/mpi-spawn-not-enough-slots-available-all-which-nodes-are-allocated-for-this-j > > <http://stackoverflow.com/questions/38087327/mpi-spawn-not-enough-slots-available-all-which-nodes-are-allocated-for-this-j> > > Thanks for your help, > Jason > _______________________________________________ > users mailing list > us...@open-mpi.org > Subscription: https://www.open-mpi.org/mailman/listinfo.cgi/users > Link to this post: > http://www.open-mpi.org/community/lists/users/2016/06/29561.php
smime.p7s
Description: S/MIME cryptographic signature