Particionamiento y Balance de Carga en Simulaciones Distribuidas de Bancos de Peces

Abstract

Partitioning and load balancing are highly important issues in distributed simulations based on spatially-explicit individual-oriented models. Choosing how to decompose the domain of an individual-oriented system and efficiently distribute the set of partitions are crucial factors at the moment to execute the distributed simulation. Furthermore, if the individual-oriented model exhibits mobility patterns, we should be able to maintain the load balancing in order to keep the global application performance as the distributed simulation evolves in time. In this work, we have developed partitioning and load balancing strategies for distributed spatially-explicit individual-oriented simulations with mobility patterns. We have used the Huth and Wissel model, which represents the coordinated and polarized motion of fish schools, in order to validate the developed strategies. Our partitioning method consists of decomposing an individual-oriented system into a collection of compact partitioning generated by means of covering radius criterion and Voronoi diagrams. The distribution model consists of grouping partitions by proximity into a set of meta-partitions. The number of meta-partitions should be equal to the number of computing resources of the distributed architecture. The dynamic load balancing strategy consists of detecting imbalance by means of a threshold-based algorithm and reconfiguration the set of meta-partitions in order to re-balance the workload. We have done exhaustive experimentation using a large set of scenarios in order to validate and verify the distributed simulator feasibility.

Date of Award	16 Jul 2012
Original language	Spanish
Supervisor	Remo Suppi Boldrito (Director)