MODELING THE PARALLEL EXECUTION OF LARGE MATRIX OPERATIONS ON A LOCALLY DISTRIBUTED COMPUTING SYSTEM.

Stafylopatis, Andreas

dc.contributor.author	Stafylopatis, Andreas	en
dc.date.accessioned	2014-03-01T02:47:48Z
dc.date.available	2014-03-01T02:47:48Z
dc.date.issued	1987	en
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/33360
dc.relation.uri	http://www.scopus.com/inward/record.url?eid=2-s2.0-0023170937&partnerID=40&md5=716e518ccb23d82ef8538ebc0e075161	en
dc.subject.other	PROBABILITY - Queueing Theory	en
dc.subject.other	ITERATIVE SCHEME	en
dc.subject.other	LARGE MATRIX OPERATIONS	en
dc.subject.other	LOCALLY DISTRIBUTED COMPUTING SYSTEM	en
dc.subject.other	MEAN ITERATION TIME	en
dc.subject.other	PARALLEL EXECUTION	en
dc.subject.other	COMPUTER SYSTEMS, DIGITAL	en
dc.title	MODELING THE PARALLEL EXECUTION OF LARGE MATRIX OPERATIONS ON A LOCALLY DISTRIBUTED COMPUTING SYSTEM.	en
heal.type	conferenceItem	en
heal.publicationDate	1987	en
heal.abstract	The evaluation of the parallel execution of general iterative schemes involving large sparse matrix operations is considered. The implementation is a locally distributed computing system including independent processors that can communicate with each other. The iterative scheme is partitioned into noninterfering tasks assigned to the processors, which proceed in parallel but can be blocked whenever information exchange between tasks is necessary. A queuing model of the system is developed for N parallel processors executing statistically identical tasks. The necessity for information exchange is expressed in a probabilistic way. The solution of the model provides performance measures, such as the probability distribution of the number of active and blocked processors, the probability of waiting, and the mean waiting time. From this parallel model a marginal model for each processor is obtained that provides the mean iteration time for the parallel scheme, therefore allowing the evaluation of the system's speedup and efficiency. 8 refs.	en
heal.publisher	IEEE, New York, NY, USA	en
heal.journalName	[No source information available]	en
dc.identifier.spage	221	en
dc.identifier.epage	224	en