How to use LSF Job management system

View a list bhosts of LSF compute nodes

 # bhostshost_name STATUS jl/u MAX njobs RUN ssusp ususp RSVfat01 OK      -0 0 0 0 0fat02 ok-16 0 0 0 0     0fat03 ok-16 0 0 0 0 0fat04 OK-              0 0 0 0 0fat05 ok-16 0 0 0 0 0fat06      Ok-16 0 0 0 0 0fat07 ok-16 0              0 0 0 0fat08 ok-16 0 0 0 0 0fat09      Ok-16 0 0 0 0 0fat10 ok-16 0 0 0 0 0 ... 

View LSF queues bqueues

To view the overall information for all queues:

# bqueuesqueue_name      PRIO STATUS          MAX jl/u jl/p jl/h njobs  PEND   RUN  suspcpu  open:active ----  2072     0  2072     0fat  open:active       -    --    -     0     0     0     0gpu  open:active       -     ---288 0   288     0mic              open:active----     0     0     0     0cpu-fat          +  Open:active----     0     0

To view information for a queue:

# bqueues Fatqueue_name      PRIO STATUS          MAX jl/u jl/p jl/h njobs  PEND   RUN  suspfat  Open:active----     0     0     0     0

View compute node load lsload

To view the overall load:

# lsloadhost_name       status  r15s   r1m  r15m   ut    pg  LS    it   tmp   SWP   memnode011             OK   0.0   0.3   0.4   0%   0.0   0 49024  193G   62G   61gnode039             OK   0.0   0.6   0.5   0%   0.0   0 49024  194G   62G   61gnode041             OK   0.0   0.0   0.0   0%   0.0   0 49024  194G   62G   61gnode050             OK   0.0   0.0   0.0   0%   0.0   0 49024  194G   62G   60gnode064             OK   0.0   0.7   0.6   0%   0.0   0 49024  194G   62G   61gnode077             OK   0.0   0.7   0.5   0%   0.0   0 49024  194G   62G   61G .....

To view the payload of a node:

# lsload node001host_name       status  r15s   r1m  r15m   ut    pg  LS    it   tmp   SWP   memnode001             OK   0.3   0.1   0.1   1%   0.0   0   332  152G   62G   61G

Submitting Jobs bsub using LSFsubmit a job manually

LSF uses BSub to submit jobs. The format of the BSub command is:

Bsub-n z-q queuename-i inputfile-o OUTPUTFILE COMMAND

Where: The Z number of threads required to submit -q the job, specifying the queue for the job submission. If you do not add an -q option, the system submits the job to the default job queue. INPUTFILEindicates that the program needs to read the file name, OUTPUTFILE indicating the output file name, the output of the job after the submission to the standard output information will be saved in this file.

For serial jobs, COMMAND You can use your program name directly. For example, the serial program is mytest submitted via LSF:

Bsub-n 1-q q_default-o mytest.out./mytest

For MPI parallel jobs, COMMAND the format is -a mpich_gm mpirun.lsf PROG_NAME . For example, a parallel program mytest , submitted by LSF, runs the job with 16 threads:

Bsub-n 16-q q_default-o mytest.out-a mpich_gm mpirun.lsf./mytest

Interactive Bulk Submissions

bsubYou can also start an interactive shell environment by committing multiple parallel jobs with the same run parameters at a time. For example, the following actions:

# bsubbsub>-N 16bsub>-Q q_defaultbsub>-o output.txtbsub> command1bsub> command2bsub> COMMAND3

Equivalent to:

Bsub-n 16-q q_default-o output.txt command1bsub-n 16-q q_default-o output.txt command2bsub-n 16-q q_default-o out Put.txt COMMAND3

writing LSF Job control Scripts

#BSUB-N 16#bsub-q q_default#bsub-o output.txt-a mpich_gm mpirun.lsf./mytest

bsubAlso accepts job descriptions from standard input. Therefore, we can write the LSF script to submit the job. BSub script is simple and easy to write, the above section of code is a bsub.script complete example named, will be redirected bsub.script through the input, submitted to LSF:

BSub < Bsub.script  

This is equivalent to:

Bsub-n 16-q q_default-o output.txt-a mpich_gm mpirun.lsf./mytest

a more complete LSF job control script

#BSUB-j hello_mpi#bsub-o job.out#bsub-e job.err#bsub-n 256source/lustre/utility/intel/composer_xe_2014.3.163/bin/ Compilervars.sh intel64source/lustre/utility/intel/mkl/bin/intel64/mklvars_intel64.shsource/lustre/utility/ intel/impi/4.1.1.036/bin64/mpivars.shmpirun= ' which MPIRun ' exe= "./mpihello" curdir= $PWDcd $CURDIRrm-F nodelist Nodes >&/dev/nulltouch Nodelisttouch nodesnp=0for host in ' Echo $LSB _mcpu_hosts |sed-e ' s//:/g ' |  Sed ' s/:n/\nn/g ' doecho $host >> nodelistecho $host | Cut-d ":"-f1 >> nodesnn= ' echo $host | Cut-d ":"-f2 ' np= ' echo $NP + $nn | BC ' Done

Other job management operationsView job Status bjobs

To check the running status of a submitted job:

Bjobs

To display the job run status in a wide format:

Bjobs-w

Show All jobs:

Bjobs-a

To display a running job:

Bjobs-r

Shows the jobs waiting to run (pending) and the reason for the wait:

Bjobs-p

Shows the cause of a suspended (suspending) job and hangs:

Bjobs-s

Show JOBID all information for this job:

Bjobs-l JOBID

terminating a bkill job

To terminate a job that is not required:

Bkill

Terminate JOBID This job:

Bkill JOBID

Remove the job directly JOBID from LSF without waiting for the job's process to end in the operating system:

Bikill JOBID

Monitor Job output bpeek

When the job is running, its standard output is displayed, and the monitor job runs:

Bpeek

View JOBID the standard output:

Bpeek JOBID

Job history Information bhist

Show the history of the job:

Bhist

Show JOBID The history of the job:

Bhist JOBID

How to use LSF Job management system