In the last 5 years, a dramatic shift has taken place in the way high-performance computation is done in the oil industry Bednar et al. (1999); Clapp and Sava (2002) and many other fields that have heavy computation requirements. No longer is a single computer with massive shared memory and multiple processors the economical choice. Instead numerous inexpensive PC based machines are clustered together for computation. This change in hardware paradigms requires a corresponding change in programing paradigms. No longer are shared memory programing models such as Open MP is sufficient. Instead a more distributed model such as Parallel Virtual Machine (PVM) or more commonly, Message Passing Interface (MPI) must be used.
A second challenge with this change in paradigm is related
to the type of problems that we deal with in the oil
industry. The volume of data makes network transfer
a non-trivial portion of the processing flow and is
the dominant element in some Mosher (1991).
In this paper, I describe the routines in SEPlib to facilitate
MPI processing and discuss some future possible directions.