Friday, 11 October 2013

Submitting jobs to a compute farm with the LSF queuing software

Submitting a job with a memory requirement
You need to use  -R "select[mem > 500] rusage[mem=500]" -M500, to request 500 Mbytes of RAM.
eg.
% bsub -o /lustre/scratch108/parasites/alc/myscript.o -e /lustre/scratch108/parasites/alc/myscript.e -R "select[mem > 500] rusage[mem=500]" -M500  /lustre/scratch108/parasites/alc/myscript

Submitting a job without a shell script
You just put the command in inverted commas at the end of the bsub command.
eg.
% bsub -R "rusage[mem=1000] select[mem>1000]" -M 1000000 "ls -al"
This runs the job 'ls -al'.

Submitting a job with a name
The -J option can be used:
eg. 
% bsub -R "rusage[mem=1000] select[mem>1000]" -M 1000000 -J "myjob50" "ls -al"

Submitting a job to run in parallel across multiple cores (in threaded mode)
If you wanted to run a job in parallel across multiple cores (in threaded mode), you need to use -n 8 
-R "select[mem > 6000] rusage[mem=6000] span[hosts=1] -M6000
, to specify that you want to run the job across 8 cores and request 6000 Mbyte of memory.
eg.
% bsub -o myscript.o -e myscript.e -n 8 -R "select[mem > 6000] rusage[mem=6000] span[hosts=1]" -M6000000 myscript

Submitting a job to the long queue
The 'normal' queue has a hard limit of 12 hours, and the 'long' queue has a hard limit of 48 hours.
To submit a job to the long queue:
% bsub -q long "ls -al"

Switching a job to the 'long' queue
To switch a job (eg. job id. 4940698) from its current queue (eg. 'normal') to the 'long' queue:
% bswitch long 4940698 

Finding out how much memory a particular queue allows
To find out how much memory the 'basement' queue allows for a job, you can type:
% bqueues -l basement
You will see a lot of information, including:
MEMLIMIT
250 G  
This tells us that the maximum memory (RAM) for a job on the basement queue is 250 Gbyte.

Killing all your jobs
For example, to kill all jobs belonging to user abc, type:
% bkill -u abc 0

Seeing which jobs are running:
% bjobs -r
This will list running jobs.
To see just pending jobs, type:
% bjobs -p 

Find out your priority for running jobs:
% bqueues -l -r normal 

If you only want a certain number of jobs to run at once:
You can make a job group eg.
% bgadd -L 40 /AvrilRepeats
Then submit jobs in that group using bsub -g /AvrilRepeats
To get information about job groups:
% bjgroup

No comments: