Skeletal Programming, Present and Future
Murray Cole
This talk considers the common theme which unites the participants,
namely that abstraction and control of structure should play an
important role in parallel programming, and suggests four related
areas in which further research is required. Firstly, in the area of
"expressivity" it is as yet unclear what might constitute the right
set of types and operators (or whether such a bounded set even exists)
and what might be the most appropriate language framework within which
to express the base level of a skeletal scheme. Secondly, the
challenge of building a tractable, compositional cost calculus for
such a model, and the choice of underlying cost/computational model
must be resolved. Thirdly, the techniques and frameworks required
for formal program derivation and transformation when the target
context is parallel must be further developed. Finally, our
implementations and concrete programming languages must be sensitive
to the needs of "real" parallel programmers, who must be convinced
that our ideas have something to offer in practice.