Question:

How do I use datascope with MPI?

Answer:

The two most common libraries for making parallel code are MPI and OpenMP. MPI is most suited for symmetric multiprocessors like IBM SP machines or, of more relevance to datascope, systems assembled from collections of many linux systems like beowolf clusters. MPI uses a message passing model for handling processing.

The warning is that because datascope is not thread safe it is not possible to run programs that do database manipulations within program segments bounded by MPI_barrier function calls. Doing so will cause chaos. Programs have to be designed to do all database manipulations by a single processor which can fundamentally change some algorithms. I'm working on a general workaround to this problem for programs that need to be run in a parallel mode, but ask me if you are interested (pavlis@indiana.edu). It is a long story.