Attaching to a Running Message-Passing Process

You can load the processes of a message-passing job into Prism.

To attach to running message-passing programs, obtain the job ID of the processes

• If you are using the LSF environment, by issuing the bjobs command (You can also get the job ID from the bsub command when it starts the job).

• If you are using the CRE environment, by issuing the mpps command (You can also get the job ID from the mprun command when it starts the job).

For example, using the LSF environment:

host4-0 54 =>bjobs
JOBID USER STAT QUEUE FROM_HOST EXEC_HOST JOB_NAME SUBMIT_TIME
15232 jay  RUN  hpc   host4-0   host4-0   chess    Sep
24 13:35 host4-1

1. Specify the job ID of the processes (not an individual process ID) on the prism command line, after the name of the executable program.

2. Use the -n (or -np, -c, -p) option when you request that Prism attach to a job. Without one of these options, Prism assumes that the ID number is a process ID rather than a job ID.

For example,

% prism -n 2 mpiprog 15232

starts MP Prism and attaches to the running processes in job 15232. See the LSF Batch User's Guide for further information about bjobs. See the Sun MPI 4.0 User's Guide: With CRE for further information about mpps.

You can also attach to a single process of a message-passing program by specifying its process ID, just as you do in scalar Prism; see " Attaching to a Running Process (Serial Program)". If you do this, however, you won't be able to view or debug what is happening in the other processes.

If you attach to a program under MP Prism, your job will be automatically detached from MP Prism if you quit or run another program. You can detach from the job by issuing the detach command from within Prism.

MP Prism only lets you detach when all the processes in the job are stopped. The detach operation itself sets them all running again, outside control of the debugger.