Toward a Flexible and Reconfigurable Distributed Simulation: A New Approach to Distributed DEVS

Abstract

With the increased demand for distributed simulation to support large-scale modeling and simulation applications, much research has focused on developing a suitable framework to support simulation across a heterogeneous computing network. Middleware based solutions have dominated this area for years, however, they lack the flexibility for model partitions and dynamic repartition due to their innate static natures. In this dissertation, a novel approach for DEVS based distributed simulation framework is proposed and implemented. The objective of such a framework is to distribute simulation entities across network nodes seamlessly without any of the commonly used middleware, as well as to support adaptive and reconfigurable simulations during run-time. This new approach, called DEVS/RMI, is proved to be well suited for complex, computationally intensive simulation applications and its flexibility in a distributed computing environment promotes a rapid development of distributed simulation applications. A hilly terrain continuous spatial model is studied to show how DEVS/RMI can easily refactor the simulations to accommodate both increases of the resolution and computation nodes. Furthermore, an agent-based valley fever model is investigated in this dissertation with particular interests on the concept of DEVS "activity". Dynamic reconfiguration of distributed simulation is then exemplified using the "activity" based model repartition in a DEVS/RMI supported environment. The flexibility and reconfigurable nature of DEVS/RMI open up further investigations into the relationship between speedup of a simulation and the partition or repartition algorithm used in a distributed simulation environment.