Re: [OMPI devel] Broken TotalView behavior in 1.5.4

2011-08-19 Thread David Gunter
Just a follow-up. I've build OMPI-1.5.4 on a non-SLURM, PBS-based system and TotalView behaves the same (incorrect) way. Still not sure what is happening. -david -- David Gunter HPC-3: Infrastructure Team Los Alamos National Laboratory On Aug 19, 2011, at 1:17 PM, David Gunter wrote: > I'v

Re: [OMPI devel] Broken TotalView behavior in 1.5.4

2011-08-19 Thread Ralph Castain
I've always had a hankering for hari kari, but I'm sure other methods would be equally acceptable :-) Suggest knocking on Nathan's door across the hall - he's been playing with that code on the trunk, and I suspect something just didn't make it across to the 1.5 branch. On Aug 19, 2011, at 1:

[OMPI devel] Broken TotalView behavior in 1.5.4

2011-08-19 Thread David Gunter
I've built the latest 1.5.4 release for one of our SLURM clusters now Totalview only appears to debug the main job launcher that calls orterun(). The actual MPI job still runs on the specified number of nodes yet TV doesn't seem to see it. Here's my launch line: $ totalview mpirun -a -n 16 my.