iRAP can be used with the LSF job scheduler with the aim of reducing the time to produce the results by splitting the analysis into multiple jobs . From the user perspective, this can be achieved without changing the parameters, input files or configuration file.
- Use the irap_lsf command (instead of irap) to run iRAP in a cluster with the LSF job scheduler. irap_lsf accepts the same parameters as iRAP but splits the analysis into multiple jobs with the aim of reducing the time to analyze the data, e.g, `irap_lsf conf=myexp.conf` `irap_lsf conf=myexp.conf mapper=tophat1 quant_method=htseq1 de_method=deseq`
- Specific parameters to irap_lsf are provided using https://wiki.archlinux.org/index.php/Environment_variables] [environment variables
| Variable | |
| MEM | Maximum memory that a job may use (in MB) |
| THREADS | Number of threads to use |
| QUEUE | Queue to where the jobs will be submitted (mandatory) |
- The variables (JOB_MEM_INCR and JOB_MAX_MEM) can be used to enable the automatic resubmission of jobs that fail due to insufficient memory requested to LSF.
For instance, lets consider that a particular run of iRAP requires 20GB. If MEM=4000 JOB_MEM_INCR=10000 JOB_MAX_MEM=16000 then irap_lsf will submit a initial job with 4GB (memory limit), then it will retry with 14GB and then irap_lsf fails (insufficient memory). If MEM=4000 JOB_MEM_INCR=10000 JOB_MAX_MEM=25000 then irap_lsf will submit a initial job with 4GB (memory limit), then it will retry with 14GB, then with 24GB (thus having enough memory). Note that in case of failure irap_lsf will resume from the point (command) that failed.