What tools are needed?
The most standard answer is that the tools must be easily applicable. It is difficult to give a general answer what is the best tool. MPI will be good in many cases. PVM may also be efficient in many cases. My impression is that both MPI and PVM have been imporoved considerably. However, further improvements are still highly desirable.
What is the future of PVM/MPI?
I think that they both have a good future (of course, unless a new tool, which is much easier to use and much more efficient, is developed). This is so because the applications become bigger and more complicated. If a mathematical model is to be used in the solution of different industrial and engineering tasks, then all involved physical and chemical processes must be adequately described. However, when this is done, the developers will normally find out that the computational problems arising after the discretization of the resulting model are enormous. Therefore the use of parallel computations becomes not only highly desirable, but also absolutely necessary. The conclusion is that better versions of MPI and PVM are required. The problem with many standards is perhaps not a very big problem: the ideas are normally the same and when you have learned two or three standards, there is not a big problem to go to the next version.
What is your vision of the
programming environment 5 years from today?
The existing big applications will become bigger, while new (and hopefully more powerful) parallel hardware will dominate the market. This will lead to an increase of the needs for standard parallel tools. It will become easier to incorporate MPI and PVM in a big application code. New intelligent compilers will take care in using automatically MPI or PVM provided that the user defines properly the parallel tasks and the domains where the input data are and the output data are to be stored.