Thomas,
Also, make sure the Debugger Tab is correctly configured: The 'Path to debugger executable' field must point to an sdm that has been compiled on your machine, and the 'Debugger session address' must be 'localhost'. There is a bug in this version that may lead to the session address being blank.
I've run the debugger on 1024 processes before, so 256 should be fine.
Greg On Feb 7, 2012, at 3:29 AM, Thomas Geenen wrote: hi Greg,
yes i can start an interactive job on multiple nodes qsub -I -l nodes=2:ppn=32 i get a $PBS_NODEFILE nodefile containing the two nodes
mpirun -np 64 hostname
shows me that it runs the job across multiple nodes
i will try the edu.illinois.ncsa.forge.pbs.interactive.openmpi resource manager do i need the dev version of ptp for that? or can i just add that resource manager to my existing ptp install?
how scalable is the debugger? will i be able to debug a job with 256 processes for instance?
thanks Thomas On Tue, Feb 7, 2012 at 1:57 AM, Greg Watson <g.watson@xxxxxxxxxxxx> wrote:
Thomas,
Are you able to run an MPI job on multiple nodes from the command line using Open MPI? You need to get this working first.
Regards, Greg On Feb 5, 2012, at 11:09 AM, Thomas Geenen wrote: i made some progress.
i start an interactive session on the cluster open a reverse ssh tunnel to my local machine and start a local openmpi resource manager. so far so good i can run and debug applications on a single node on the cluster.
However i cant figure out how to run on multiple nodes. when i start a parallel debug session ptp writes a routing_file file in my run dir that contains only references to the current node.
i seems that i cannot provide my own host file.
thanks Thomas On Fri, Feb 3, 2012 at 3:40 PM, Thomas Geenen <geenen@xxxxxxxxx> wrote:
dear all,
i try to run a parallel debug session through the interactive pbs resource manager.
launching a parallel debug session with the interactive resource manager results in an error message
Configuration is missing run commands for job submission <some number> debug
i found two post in the mailing list dealing with this same issue but no solution
i can run jobs through this resource manager without a problem and i can also start a remote parallel debugging session using the openmpi resource manager.
some system info (Open MPI) 1.4.2 org.eclipse.ptp.linux.x86_64_5.0.4.201111121445/
thanks Thomas
_______________________________________________
ptp-user mailing list
ptp-user@xxxxxxxxxxx
https://dev.eclipse.org/mailman/listinfo/ptp-user
_______________________________________________ ptp-user mailing list ptp-user@xxxxxxxxxxx https://dev.eclipse.org/mailman/listinfo/ptp-user
|