Show simple item record

dc.contributor.authorScott, Robert B.en
dc.date.accessioned2019-02-15T14:20:53Z
dc.date.available2019-02-15T14:20:53Z
dc.date.issued1994
dc.identifier.urihttp://hdl.handle.net/1842/33864
dc.description.abstracten
dc.description.abstractThere has been much research interest in efficient implementations of the Committed Choice Non-Deterministic (CCND) logic languages on parallel computers. To take full advantage of the speed gains of parallel computers, methods need to be found to automatically distribute goals over the machine processors, ideally with as little involvement from the user as possible.en
dc.description.abstractIn this thesis we explore some automatic goal distribution strategies for the execu¬ tion of the CCND languages on commercially available distributed memory parallel computers.en
dc.description.abstractThere are two facets to the goal distribution strategies we have chosen to explore:en
dc.description.abstractDEMAND DRIVEN: An idle processor requests work from other processors. We describe two strategies in this class: one in which an idle processor asks only neighbouring processors for spare work, the nearest-neighbour strategy; and one where an idle processor may ask any other processor in the machine for spare work, the allprocessors strategy.en
dc.description.abstractWEIGHTS: Using a program analysis technique devised by Tick, weights are attached to goals; the weights can be used to order the goals so that they can be executed and distributed out in weighted order, possibly increasing performance.en
dc.description.abstractWe describe a framework in which to implement and analyse goal distribution strategies, and then go on to describe experiments with demand driven strategies, both with and without weights. The experiments were made using two of our own implementations of Flat Guarded Horn Clauses — an interpreter and a WAM-like system — executing on a MEIKO T800 Transputer Array configured in a 2-D mesh topology.en
dc.description.abstractAnalysis of the results show that the all-processors strategies are promising (AP-NW), adding weights had little positive effect on performance, and that nearest-neighbours strategies can reduce performance due to bad load balancing.en
dc.description.abstractWe also describe some preliminary experiments for a variant of the AP-NW strategy: goals which suspend on one variable are sent to the processor that controls that variable, the processes-to-data strategy. And we briefly look at some preliminary results of executing programs on large numbers of processors (> 30).en
dc.publisherThe University of Edinburghen
dc.relation.ispartofAnnexe Thesis Digitisation Project 2019 Block 22en
dc.relation.isreferencedbyAlready catalogueden
dc.titleAutomatic goal distribution strategies for the execution of committed choice logic languages on distributed memory parallel computersen
dc.typeThesis or Dissertationen
dc.type.qualificationlevelDoctoralen
dc.type.qualificationnamePhD Doctor of Philosophyen


Files in this item

This item appears in the following Collection(s)

Show simple item record