Since moving to Indiana University with my advisor, I have made frequent use of the HPC resources made available to grad students, staff, and faculty here. I have made particularly frequent use of the Mason cluster, which has 16 machines, each with 500 GB available memory and 32 processors.
However, over this last weekend I wanted to make use of IU’s new Big Red II. This is your much more typical HPC environment designed for distributed-memory MPI jobs utilizing processors on different machines in parallel. I occasionally run MPI jobs, so I am familiar with the likes of
mpiexec (Big Red II uses
aprun). These particular jobs, however, were not MPI jobs, and therefore needed to be run in “cluster compatibility mode”, or CCM for short. Below is a simple example launch script that will execute a BLAST job in CCM mode on Big Red II.
#!/bin/bash #PBS -N NameOfYourBlastJob # a name for the job #PBS -l nodes=1:ppn=32 # request a single machine with 32 processors for this job #PBS -l walltime=12:00:00 # request a 12-hour time slot for this job #PBS -q cpu # use the 'cpu' queue #PBS -l gres=ccm # use Cluster Compatibility Mode #PBS -M email@example.com # email address to which notifications will be sent #PBS -m bea # send email notifications when the job begins, ends, and/or aborts # Place 'module load ccm' in your ~/.modules file WD=/N/dc2/scratch/username/working-directory ccmrun blastx -query $WD/sequences.fna \ -db $WD/reference-proteins.faa \ -out $WD/blast-results.xml \ -evalue 1e-15 \ -outfmt 5 \ -show_gis \ -num_alignments 10 \ -num_threads 32
Then as usual, the job can be launched by running