Implementation and evaluation of OR parallel Prolog models on distributed systems
V. Benjumea, M. Roldan, J.M. Troya.
Abstract
A Prolog parallel interpreter with which to evaluate several
approaches in distributed OR parallel implementations is shown. Two
execution models based on a process tree, which differ mainly in the
process granularity, are considered. Furthermore, two memory
management strategies based on a closed environment are examined. The
evaluation method consists of comparing real results obtained after
running four different versions which combine the different issues.
The interpreter has been implemented on a 16 Transputers Parsys
Supernode, using Occam as the development language. The process and
memory management have been implemented in a dynamic way, allowing
improvement of the system's efficiency and reduction in memory
waste. A dynamic load balancing mechanism has been implemented. Some
statistics made using the interpreter show a speed-up of 3.75 for 4
processors and 12.57 for 16 processors.
Keywords: Parallel interpreter, logic programming, AND parallelism, OR
parallelism, distributed memory systems, closed environment memory
model