Hi everyone!
I'm having a slight problem running LAM-MPI jobs under PBSpro. The
processes/nodes ranking higher than 0 respond with an error when they
attempt to start:
user not in group
I'm currently talking to the guys at PBSpro, and they are wondering
whether LAM for some reason could change the user/group id when it
spawns processes (does LAM spawn any processes? I believed PBS starts 1
proces on each node and I don't specify lamd). Authentication is by
PublicKey, the cluster uses NIS, and the user's working directory is nfs
mounted. LAM 7.0.2 is configured to employ PBSpro's tm. Platform is RH
8.0. Jobs behave excellently when run manually, i.e. lamboot; mpirun.
Best regards, Jess Michelsen
|