Some Essential Side-effects of Prolog on a Distributed Implementation.
Lourdes Araujo
Computers and Artificial Intelligence, 19(2), (2000), p. 147-168.

This work describes an implementation of some essential side-effects
of Prolog: cut, findall, and input/output, on a
distributed memory system. The techniques designed are advantageous
when applied to parallel systems based on recomputation, such as PDP
(Prolog Distributed Processor), a model for Independent_AND/OR parallel
execution of Prolog. Nevertheless, they are also valid for any
distributed memory system. The implementation of the cut predicate
exploits as much parallelism as possible, including the parallel
computation of branches of the search tree which depend on a cut.
However, the computation of a branch which does not depend on a cut
is never delayed to control computations depending on a cut.
The model proposed for the findall predicate reduces the
communication as much as possible by distributing the control of the
execution of findall, and takes advantage of the success paths
used in the recomputation to perform the sorting of the solutions.
The PDP model for the exploitation of combined parallelism has allowed a
straightforward introduction of the input/output predicates.
Results obtained with an implementation of the models on a network of
workstations by means of the PVM (Parallel Virtual Machine) software
package are presented.