Future Work

• Study task assignment techniques based on the combination of granularity information and scheduling techniques such as those used in data parallelism.

• Combine with Partial Evaluation Techniques in order to create coarse granularity tasks (e.g. by means of unfolding operations, loop unrolling, etc.)

• Apply the results of these tools to the version of parallel ECLPS running on a network of workstations and the distributed version of Ciao-Prolog.

• Improve cost analysis for ``divide and conquer'' programs: inference of size relations between arguments.
• Extend to Concurrent Constraint Languages.
• Study average case analysis:
  • Giving an acceptable definition of ``average'' requires defining a probability distribution on the possible inputs, and this seems nontrivial.
  • However, one can imagine that profiling techniques might be usable for estimating input distributions.