Interactive Job Execution with LoadLeveler


While developing applications it may be necessary to execute a job under interactive control. llrun creates and submits a LoadLeveler job under the direct control of the user's shell.

Using llrun


llrun [<llrun_options>] [<mpirun_options>] [<executable>]

llrun can be used with the same command line options like mpirun. There are only a few additional parameters.

llrun options:

  -B             submit batch job
  -w  <hh:mm:ss> submit batch job with wallclock limit  (Default: 00:30:00)
  -o  <filename> do not run/submit job but save to file 
  -q             do not queue interactive job if insufficient resources
  -v             verbose
  -h -?          help (this message)

mpirun options:

  -h                            Provides this extended help information.
                                Can also use '-help'.
  -version                      Display version information.
                                Can also use '-v'.
  -np <compute_nodes>           The number of compute nodes to use for the job.
                                Can also use '-n' or '-nodes'.
  -mode <CO or VN>              Execution mode: COprocessor or VirtualNode mode.  Default is CO.
  -exe <binary>                 Full path of the binary to execute.
  -cwd <directory>              Current working directory of the job,
                                as seen by the compute nodes.
                                Can also use '-wdir'.
  -host <service_node>          Hostname of the service node.
                                Overrides MMCS_SERVER_IP environment variable.
  -shell <remote_shell>         Remote shell mechanism to use.
                                Defaults to RSH, SSH is available.
  -env <"<env vars>">           Environment variables that should be set.
                                (Must be enclosed in double quotes).
  -exp_env <env vars>           Environment variables in the current environment to export.
  -mapfile <mapfile>            File containing a user specified MPI topology
  -args <"<arguments>">         Arguments to pass to the job.
                                (Must be enclosed in double quotes).
  -timeout <seconds>            The limit of the job execution time in seconds.
  -start_gdbserver <gdbserver>  Start gdbserver for job.  Must specify full path to gdbserver.
  -label                        Add labels (STDOUT, STDERR, and MPI Rank) to the job output.
  -nw                           Return MPIRUN job cycle status instead of the job exit status.
  -noallocate                   Do not allocate the partition.  The job will only start
                                if the partition was already in the Initialized state.
  -backend <binary>             Use a specific mpirun back-end on the service node.
  -kernel_options <options>     Specify kernel verbose options.
  -verbose <0 - 4>              Verbosity level of mpirun.  Default is 0.
  -trace <tracelevel>           Create a detailed trace as mpirun executes.
                                Output is to a a file in the current directory.
  -only_test_protocol           Used to test the mpirun front-end/back-end communications.
                                No actual job will be run.
  -strace <all|none|n>          Enable syscall trace for all nodes, no nodes, or node n
                                where n is an MPI rank number
  -shape <NxNxN>                Shape of the job in XxYxZ format.
                                If not specified you must use -np.
  -psets_per_bp <n>             Number of PSets per BP required in the partition.
  -connect <TORUS or MESH>      Compute node connections.  Default is MESH.
  -enable_tty_reporting         Correctly report tty status to the control system

The options -free and -partition are ignored.
last change 23.02.2007 | Michael Stephan | Print