Divided We Stand: Parallel Distributed Stack Memory Management

Abstract

We present an overview of the stack-based memory management techniques that we used in our non-deterministic and-parallel Prolog systems: &-Prolog and DASWAM. We believe that the problems associated with non-deterministic and-parallel systems are more general than those encountered in or-parallel and deterministic and-parallel systems, which can be seen as subsets of this more general case. We develop on the previously proposed "marker scheme", lifting some of the restrictions associated with the selection of goals while keeping (virtual) memory consumption down. We also review some of the other problems associated with the stack-based management scheme, such as handling of forward and backward execution, cut, and roll-backs.

Parallel execution of prolog programs

“…Hermenegildo [1986a] showed the relationship between scheduling and memory management, and provided ideas on using more sophisticated scheduling techniques for guaranteeing a better match between the logical organization of the computation and its physical distribution on the stacks-with the aim of simplifying backtracking and memory performance. This issue has been studied further in [Shen and Hermenegildo 1994;1996a], where flexible related scheduling and memory management approaches are studied. Related research on scheduling for independent and-parallel systems has also been proposed by Dutra [1994].…”

“…A further distinction has been made in the literature [Pontelli et al 1996;Shen and Hermenegildo 1994], regarding how outside backtracking is carried out: -private backtracking: each worker is allowed to backtrack only on the computations lying in their own stacks. Thus, if backtracking has to be propagated to a subgoal lying in the stack of another worker P , then a specific message has be sent to P , and P will (typically asynchronously) carry out the backtracking activity; -public backtracking: each worker is allowed to backtrack on any computation, independently from where it resides-it can also backtrack on computations lying on the stack of a different workers.…”

“…Markers are used to delimit stack sections (horizontal cuts through the stack set of a given abstract machine, corresponding to the execution of different parallel goals) and they implement the storage recovery mechanisms during backtracking of parallel goals in a similar manner to choice points for sequential goals [Hermenegildo 1986b;Shen and Hermenegildo 1994;1996a]. As illustrated in Fig.…”

“…Also, agents are started and put to "sleep" as needed in order not to overload the system when no parallel work is available. Several scheduling and memory management strategies have been studied for the &-Prolog system [Hermenegildo and Greene 1991;Shen and Hermenegildo 1994].…”

“…DASWAM has been designed as an extension of the PWAM model used for independent andparallelism. Memory management is analogous to PWAM-and relies on the use of parcall frames to represent parallel conjunctions, and on the use of markers to delimit segments of stacks associated with the execution of a given subgoal [Shen and Hermenegildo 1994].…”