After completing the installation, configuring Myrinet MPICH (if you installed the gm-mpich package), and installing the gm driver, you can run the confidence script to verify that the installation process has completed successfully.
The confidence script checks the following:
That the user can use the rsh(1) command among all nodes (in any combination) within the cluster
That the current working directory is accessible on all nodes
If gm-mpich was installed, that the gm driver is installed on all nodes
If gm-mpich was installed, that the gm configuration file exists on all nodes
That all nodes are available to PBS
That the user can submit a PBS request and get output back
If mpich was installed, that MPICH's tstmachines(1) command can be run successfully
That a user can run a simple MPI job that uses all execution nodes in the cluster
That a user can submit a PBS request that will run a simple MPI job that uses all execution nodes in the cluster
You must run confidence on the head node.
Log in to the head node.
Change to a directory that is NFS mounted to all nodes in the cluster and that has an identical path on each node. For example:
# cd /nfs_shared/tmp
Note: The confidence test of MPICH requires an NFS mounted directory. If you do not have such a directory, you can use the -n option to the confidence script. However, if you use this option, MPICH's tstmachines(1) script will not be executed.
Execute the confidence script.
The following example shows output for a successful test on a two-node cluster when confidence is run from an NFS mounted directory:
[root@ace08 ~]# cd /nfs_shared/tmp [root@ace07 tmp]# /usr/local/ace/bin/confidence Testing that user has rsh access to and from all hosts. Shipping a script to each host that will rsh to all other hosts. PASS All hosts can rsh to every other host. INFO hostnames match ACE node names. INFO Current Working Dir is NFS mounted on all hosts. PASS All 2 nodes configurated and available to PBS. Testing that PBS can execute a simple script ... Adjusting PBS configuration to allow root to submit requests. Submitting test job to default PBS queue. qsub output: 0.ace08 Waiting up to 60 seconds for PBS request to finish ... PASS PBS request has completed. Checking that output matches cluster node list ... PASS PBS qsub output contains node ace07 output. Restoring PBS acl_roots configuration. Basic PBS test complete. Assuming /nfs_shared/tmp/ace-conf.13221 is NFS accessible on all nodes. Testing MPICH communication using tstmachines ... PASS MPICH's tstmachines returned successfully. Testing that you can run an mpitst program. Compiling mpi test binary mpitst. Running mpich/gm_mpich /nfs_shared/tmp/ace-conf.13221/mpitst test program ... PASS mpitst test program ran on all 2 nodes. Hello World! ace07 is 1 of 2 Hello World! ace08 is 0 of 2 Basic mpi test complete. Testing that an mpi program can be run via PBS ... Adjusting PBS configuration to allow root to submit requests. Submitting mpi job to default PBS queue. qsub output: 1.ace08 Waiting up to 60 seconds for PBS request to finish ... PASS The submitted mpitst test program ran on all 2 nodes. Restoring PBS acl_roots configuration. confidence: No errors detected. Success!!! |
To get help for confidence, use the -h option. If you want debug information, use the -v option.