Parallel applications are inherently more difficult to develop and tune then serial applications. This difficulty is compounded by a lack of tools for developing parallel programs. The most important development tools are a parallel debugger and performance analysis tools, including tools for message trace visualization. A parallel debugger should not require a separate window for each process, and should understand and be able to display MPI message queues.
Tools are separate from the MPI implementation itself, and are treated only superficially in this report.