Abstract

. This paper considers the issue of dynamic task control in the context of a parallel Haskell implementation on the GRIP multiprocessor. For the first time, we report the effect of our task control strategies on task granularity, as measured in terms of dynamic heap allocations. This gives a concrete means of measuring the effectiveness of these strategies other than wall-clock timings, which are notoriously uninformative. 1 Introduction Compared with most conventional approaches to parallel programming, it is comparatively easy to write parallel functional programs. Deadlock is impossible. Because programs are deterministic, they may be debugged on sequential hardware. Programs can be highly scalable, running without change on different processor configurations or even architectures. However this flexibility places a high load on the functional compiler and runtime system. Our approach is to exploit parallelism which is notionally fairly fine-grained, but which can be dynamically com...

Citations