OpenPBS, Torque and MPI

Let us say you want to run BLAST for 25,000 predicted genes against the NCBI nr protein database to determine potential homologies. If running this calculation on one computer takes 100 hours (>4 days), doing it in parallel in a cluster of 100 nodes would take only 1 hour. Most bioinformaticians are familiar with this type of parallel computation.

When multiple computers are connected as a cluster, it is helpful to have a set of programs to submit jobs to those computers. PBS and Torque does that.

Those computers can run jobs independently, or share jobs. MPI or message passing interface, is helpful for implementing shared jobs. A more advanced application of MPI uses message passing feature of the architecture, and has been utilized by parallel genome assemblers such as ABySS.

MPI can run in client-server mode.

MPI uses TCP-IP. So, it is good for computer to computer, but can also work on single computer.