next up previous
Next: Data Parallel Fortran Up: No Title Previous: The Best Architecture?

Software

The adoption of parallel machines as a mainstream computing tool is held up by the lack of application codes that can run on them. Most successful uses of parallel machines have come from academic and research applications with less than 10,000 lines of code and where the software and parallel algorithm have been developed from scratch. The task of reimplementing large codes, such as the important 100,000 to over 1,000,000 line industrial applications for parallel machines, is highly nontrivial.

The need to rework the software is clearly the major inhibitor to the rapid adoption of parallel machines. However, there has been significant progress in developing ``portable scalable languages and software environments.'' These allow us to reimplement or develop new applications with the assurance that the resultant software will run well on all the current and projected parallel machines for which the problem is suitable.

There are no compelling new ``parallel languages,'' but rather the successful approaches have extended existing languages. We will briefly discuss

Fortran here, but similar remarks can also be made for C, C++, Ada, Lisp, etc. There are two classes of extensions to Fortran, which we discuss in turn.



Geoffrey Fox, Northeast Parallel Architectures Center at Syracuse University, gcf@npac.syr.edu