Profiling Parallel Functional Computations (Without Parallel Machines)

Abstract

Pick up a textbook on functional programming. Somewhere amidst the introductory remarks the author will almost certainly claim that functional languages are well suited for programming parallel computers. But are they? Not, it would seem, without the use of annotations to indicate where parallel evaluation should be performed. Such annotations have at least two advantages over a laissez-faire approach to parallelism. Firstly, they make it easier to reason about the program’s behaviour when it is run on a parallel machine. The annotated program describes both what is to be computed and how it is to be done, while maintaining a clean separation between the two aspects. Secondly, annotations mean that the compiler does not have to tackle the difficult problem of determining a suitable parallel evaluation strategy for the program. That remains the programmer’s responsibility.