High Performance Computing

Parallel Environment

The core of the parallel communications system on the cluster is the Message Passing Interface (MPI). When it comes to running your parallel code it is very necessary that the user understands that it is an internode communications system, over a network (gigabit), that will necessarily come with a communications overhead.

The vanilla communications system just uses MPICH. Streamline have installed an extra layer (PM) to decrease inter-node communications time; the result is SCore (SCore = MPICH + PM). You do not need to modify our code to take advantage of this: you just need to compile your code against the SCore libraries.

LAM-MPI uses MPI in the users coding but, but again the user would compile against the LAM-MPI libraries.