Assessing and Comparing the Usability of Parallel Programming Systems

Parallel programming is widely acknowledged to be more difficult than sequential programming. One reason for this is that parallel programming systems are more difficult to use than their sequential counterparts. In particular, few parallel programming systems can support the software engineering requirements of large applications. We intend to assess and compare the usability of a variety of parallel programming systems using a small suite of chained applications called the Cowichan Problems. 1 Introduction In their 1989 survey paper, Bal et al. listed more than 300 parallel programming systems [4]. A similar survey today could well uncover twice as many, but a poll of application programmers would probably find that the only ones in general use are data-parallel extensions to Fortran and procedural message-passing libraries of various flavours. It is therefore interesting to ask: why have so many parallel programming systems been developed, and why are most of them used so little? ..