Launches parallel tasks on a set of hosts.
Synopsisblaunch [-n] [-u host_file | -z host_name ... | host_name] [-use-login-shell | -no-shell] command [argument ... ]
Most MPI implementations and many distributed applications use the rsh and ssh commands as their task launching mechanism. The blaunch command provides a drop-in replacement for the rsh and ssh commands as a transparent method for launching parallel applications within LSF.
The blaunch command supports the following core command line options as rsh and ssh:
- rsh host_name command
- ssh host_name command
All other rsh and ssh command options are silently ignored.
The blaunch command transparently connects directly to the RES and sbatchd daemons on the remote host. Remote tasks are created and tracked, and the connection back to LSF is maintained. You do not need to insert the pam or taskstarter commands, or any other wrapper.
The blaunch command works only under LSF. It can be used only to launch tasks on remote hosts that are part of a job allocation. It cannot be used as a stand-alone command.
When no host names are specified, LSF runs tasks on all allocated hosts, one remote task per job slot.
- Only the following signals are supported: SIGKILL, SIGSTOP, SIGCONT.
- The -n option is not supported.
- The CMD.EXE /C <user_command_line> command is used as intermediate command shell when the -no-shell option is not specified.
- The CMD.EXE /C command is not used when the -no-shell option is specified.
- Windows User Account Control must be configured correctly to run jobs.
- Any tasks killed outside of LSF (for example, with the taskkill command) are assumed to have a normal exit.
For parallel jobs launched with the blaunch command, job control actions that are defined in the JOB_CONTROLS parameter in the lsb.queues file take effect only on the first execution host. Job control actions that are defined in the queue do not affect tasks that are running on other hosts. If the JOB_CONTROLS parameter is defined, the default job control signals of LSF (SUSPEND, RESUME, TERMINATE) do not reach each task on each execution host.
For parallel jobs launched with the blaunch command, automatic job resizing is a signaling mechanism only. It does not expand the extent of the original job that is launched with the blaunch command. The resize notification script is required along with a signal-listening script. The signal-listening script runs more blaunch commands on notification to allocate the resized resources to make them available to the job tasks. For help creating signal listening and notification scripts, contact IBM Support.
- Standard input is taken from /dev/null. Not supported on Windows.
- -u host_file
- Runs the task on all hosts that are listed in the host_file file.
Specify the path to a file that contains a list of host names. Each host name must be listed on a separate line in the host list file.
This option is exclusive of the -z option.
- The name of the host where remote tasks are to be launched.
- -z host_name ...
- Runs the task on all specified hosts.
The host name value for rsh and ssh is typically a single host name, but you can use the -z option to specify a space-delimited list of hosts where tasks are started in parallel.
Specify a list of hosts on which to run the task. If multiple host names are specified, the host names must be enclosed by quotation marks (" or ') and separated by white space.
This option is exclusive of the -u option.
- Launches commands through user's login shell.
Applies only to UNIX and Linux hosts.
- Launches commands without any intermediate shell.
- command [argument ...]
- Specify the command to run. The command must be the last argument on the command line.
- Prints command usage to stderr and exits.
- Prints LSF release version to stderr and exits.
Exit status is 0 if all commands are run correctly.
lsb_getalloc, lsb_launch APIs