Performance and scalability in distributed cluster-based individual-oriented fish school simulations

• Autores: Francisco Borges, Roberto Solar, Remo Suppi Boldrito, Emilio Luque
• Localización: Actas de las XXIV Jornadas de Paralelismo / coord. por Guillermo Botella, Alberto A. del Barrio García, 2013, ISBN 978-84-695-8330-2, págs. 401-406
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • With increasing computational powerand parallel/distributed computing it is possible tocreate more complex models to achieve simulationresults closest to reality. More realistic results allowscientists to reach conclusions and gain knowledgeabout the system under study. However this realisticsolutions implies increasing the complexity ofmodel, consequently it’s required more communication.The problem is that communication has negativeimpact in parallel distributed simulations. Two of themain challenge in distributed simulation is how to distributeindividuals efficiently through distributed architectureand the efficiency of communication strategies.In this work we compare our cluster-based partitioningapproach based on Voronoi diagrams andcovering radius criterion by using asynchronous andsynchronous communication strategy for distributedindividual-oriented fish school simulation. The maincontributions are: we showed that cluster-based partitioningapproach is scalable in both communicationstrategies and show that asynchronous communicationstrategy has best speedup in comparison to synchronouscommunication strategy.