Cost-based partitioning for distributed simulation of hierarchical modular DEVS models

Abstract

The main objective of this research is to design and implement a class of generic partitioning algorithms for hierarchical, modular Discrete Event Specification System (DEVS) models for distributed simulation. To attain the goal of this dissertation, a set of partitioning algorithms is designed using the cost analysis methodology. For more than a decade, abundant research has been conducted to develop partitioning algorithms that can find optimal, or reasonably acceptable, solutions for various partitioning problems. These employ methods such as simulated annealing, random partitioning, heuristic partitioning, and hierarchical clustering. In this dissertation, a new Generic Model Partitioning (GMP) algorithm for hierarchical modular DEVS models is proposed for distributed simulation. The GMP algorithm decomposes a given hierarchical model into a set of partition blocks and provides reasonable solutions for distinct partitioning problems based on a cost analysis methodology. It also minimizes model decomposition during the partitioning process and guarantees incremental quality of partitioning (QoP) improvements until a best partitioning is attained. A series of cost measures for cost generation, cost evaluation, and cost aggregation are introduced. Since a cost measure is a parametric method, subject to certain axioms, the proposed algorithm is generic and applicable any family of models provided there is a way to manipulate the appropriate cost information. A class of advanced algorithms derived from the GMP algorithm is also presented to tackle sophisticated issues associated with various distributed system configurations.