Specifies the hard limit, soft limit, or both limits for the CPU time used by all processes of a serial job step. For example, if a job step runs as multiple processes, the total CPU time consumed by all processes is added and controlled by this limit.
For parallel job steps, LoadLeveler enforces these limits differently. Parallel job steps usually have tasks running on several different nodes and each task can have several processes associated with it. In addition, the parallel tasks running on a node are descendants of a LoadL_starter process. Therefore, if you specify a hard or soft CPU time limit of S seconds and if a LoadL_starter has N tasks running under it, then all tasks associated with that LoadL_starter will be terminated if the total CPU time of the LoadL_starter process and its children is greater than S*N seconds.
If several LoadL_starter processes are involved in running a parallel job step, then LoadLeveler enforces the limits associated with the job_cpu_limit keyword independently for each LoadL_starter. LoadLeveler determines how often to check the job_cpu_limit by looking at the values for JOB_LIMIT_POLICY and JOB_ACCT_Q_POLICY. The smaller value associated with these two configuration keywords sets the interval for checking the job_cpu_limit. For more information on JOB_LIMIT_POLICY and JOB_ACCT_Q_POLICY see Collecting job resource data on serial and parallel jobs.
The syntax is:
job_cpu_limit = hardlimit,softlimit
For example:
job_cpu_limit = 12:56,12:50
See Limit keywords for more information on the values and units you can use with this keyword.