Skip to content

Lesson 3: Submitting Jobs Introduction

All jobs run on the HPC systems should be submitted via the queuing system, Grid Engine.

The two most common ways to submit a job are:

  • Batch processing - computation of a set of commands without manual intervention
  • Interactive processing - computation with direct user input into a terminal on a compute node

Upon submission, your job will be assigned a unique job number (JID) which can be used for post-submission commands, described later in these tutorials.

The output of your job will be contained in the 'example.oJID' file, unless overridden by the -N job resource (described below).

Resource Specification

Job resources define the allocation of physical hardware and describe how the job should run. They are specified at the top of the file using the #$ <resource> syntax. The main resources to be requested on the cluster are cores, memory and runtime. For more information regarding job resources, see here.

A few common resources are as following:

  • -cwd - instruct the scheduler to start the job in the current working directory
  • -j y - join the output files (stdout and stderr) into a single file
  • -pe PE X - request X number of cores within a specific PE (parallel environment)
  • -l h_rt=h:m:s - the wallclock runtime of your job, in h:m:s format -l h_vmem - he amount of memory your job requires per core in Gigabytes
  • -N <job_name> - specify a custom job name, output files will be in the format 'job_name.oJID'

The majority of job resources are optional however, if the runtime and memory requests are omitted, the default of 1 hour and 1GB is requested, respectively.