LSF Cheat Sheet
LSF is a software package that enables a group of independent computers to be harnessed as one coherent computing machine. It distributes computing work across the various computers in a consistent and managed manner, allowing for maximum utilization of cpu, memory, and other compute resources such as software licenses.
The following are the common LSF commands. The larger ones are more commonly used
bsub
This command is probably the most important command in LSF. Use this command to submit your job(s) to the LSF system. There are many options to this command but in most situations, only a few are really needed. Typically all that is needed, is to prepend the command or script you normally execute with "bsub". Once you do this, your job will be submitted to the LSF system. Your job may start running in a matter of seconds, or may stay pending for several minutes before dispatched. Factors which effect job pending time include the LSF queue configuration, as well as other jobs currently in the LSF system submitted by other users.
- bsub -q priority myscript -> send myscript to the priority queue
- bsub -q normal -m ibmblade1 myscript -> send myscript to the priority queue run on host ibmblade1.
- bsub -R "rusage[mem=1000]" -o jobout myscript -> send myscript to default queue configured, requests 1G of memory for the duration of the job, and send the job's stdout and stderr to the file called "jobout". Note, if -o is not used, or the job does not write stdout/stderr to a file itself, you will by default, receive the entire job output via email.
- The following are examples of various bsub job submission commands:
- The following are examples of various bsub job submission commands:
bjobs
This command replies back with a list of your jobs currently active in the system. This includes jobs that are pending (waiting to be disptached for execution) and those executing. You can get a list of all the jobs currently in the LSF system by executing "bjobs -uall". To get details on your jobs specify the -l option, and you can look at a particular job by specifying the jobid like this "bjobs -l 45322".
bkill
This command kills your job(s). You can specify one or more jobs to kill, via the jobid. You can kill all your jobs if you specify a zero, like "bkill 0".
bhist
This command replies back with a list of your jobs that have finished more than 1 hour ago. For jobs that have finished within the last hour, use the bjobs command above.
bpeek
This command replies back with the stdout (standard output) of your job. Bascially you can "peek" at the output of your job while its running, to see whats happening. Note, if you specify the -o option to the bsub command when you submit your job, bpeek will not return any output, as the -o option directs the job output to a file. Please inspect the output file in that case.
bqueues
This command gives you information about which queues exist on the cluster. bqueues -l blah will give you detailed information about the blah queue, for example, how long jobs in that queue can last.
lsid
This command replies back with the cluster name and the master host. This command verifies that you are connected to the LSF cluster.
lsload
This command replies back with the current loading of the hosts in the cluster. It lists the various host load parameters and their values at the time the lsload command is executed.
bhosts
This command replies back with a list of hosts that are part of the LSF system. Various parameters such as host status, MAX number of job that be executed on the host at any one time, number of jobs (NJOBS) dispatched to the host, and the number of jobs currently running (RUN) on the host are shown.
