|
Running Jobs on the Bauer Core Linux ClusterOnce you have an account on the cluster and know how to log in and transfer your files, you are ready to run jobs. Submitting jobsRather than running a job from the command line (by typing something like “blastall -p blastp -i infile -o outfile -d nr”), you need to submit a job to the LSF queueing system. The system will wait until a processor is available to run your job. It will run the job on a different computer (not portal) and return the results to you as if you had run them on portal. This may sound complicated, but all you need to do to submit a job is type the word "bsub" before the command you were going to type anyway. For example, "bsub blastall -p blastp -i infile -o outfile -d nr". QueuesIn order to ensure fair usage of the cluster, there are a number of queues (waiting lines) for jobs. When you submit a job with the bsub command, it will by default go into the normal queue. Alternatively, you can specify a queue: "bsub -q long bsub blastall -p blastp -i infile -o outfile -d nr". Most queues have a time limit. Any job running for longer than the queue’s time limit will be killed automatically by LSF.It is usually a good idea to run a small part of your job and see how long it takes, in order to figure out to which queue you should submit your job. Also, some queues can be used by only some users. "bqueues -ufred" will say which queues user fred can use. Finally, each queue has a priority, so it may be able to suspend jobs in some other queues. (On average, a shorter job will interrupt a longer one; when the shorter job is done, the longer one will resume.) Here is a summary of queues that are available.This information (and more) can also be accessed with "bqueues -l".
Other LSF commandsFor more information on bsub, bqueues, and other LSF commands, read the LSF cheat sheet. Bauer Cluster Support: cluster_admin@cgr.harvard.edu
|