> Am 05.10.2016 um 10:10 schrieb Mark Dixon <m.c.di...@leeds.ac.uk>: > > On Tue, 4 Oct 2016, Reuti wrote: > ... >> Yeah, I had the idea of different temporary directories some time ago, as >> some applications like Molcas need a persistent one across several `mpiruns` >> on each node and how to delete them again. > ... > > Hi Reuti, > > Interesting! > > I wonder what happens with $TMPDIR cleanup if you have two concurrent 'qrsh > -inherit' processes, targeting a single queue instance, that exit at > different times. For example, can the execd be fooled into keeping the > slave's $TMPDIR throughout by starting a blocking dummy qrsh before the > series of mpirun's?
I would say this should work too. But one has to take care that one additional `qrsh -inherit ...` call per node must be allowed. Nowadays where many MPI implementations use only one such call per node to start a daemon and start threads it might be easier to have one free-qrsh-per-node than in former MPICH(1) times where each process needed a qrsh call. -- Reuti _______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
