Exploiting the Concept of Activity for Dynamic Reconfiguration of Distributed Simulation

Abstract

In this paper, we specialize the concept of "activity", defined in earlier work, to measure the heterogeneity of model behavior using the temporal- spatial distribution of its local transitions in a 2D cellular space. We then show how to employ this "activity" metric to balance the computation load using the dynamic reconfiguration of the distributed simulation. We also show how the degree of improvement depends on the heterogeneity of the activity's distribution. That is, high concentrations of activity in space that change relatively slowly during simulation can be exploited to reduce execution time significantly within an appropriate infrastructure for dynamic reconfiguration in a DEVS based distributed simulation framework. In contrast to other dynamic load balancing approaches, the activity-based approach discussed here exploits model properties directly rather than relying on resource-based measurements as the basis for its reconfigurations.