A Dynamic, Distributed, Hierarchical Load Balancing for HLA-Based Simulations on Large-Scale Environments

The dynamic management of load in large-scale distributed systems is essential for the performance of simulations due to the influence that computing capacity and work load have on execution time. The High Level Architecture (HLA) was designed with the purpose of providing management services in order to organize distributed simulations, but the framework does not offer tools for controlling load imbalances of distributed simulations. In order to provide a generic solution for the simulation load imbalances, many approaches have been proposed. These schemes are limited to solve balancing issues regarding specific simulation or environment characteristics. With focus on balancing the computational load specially for HLA-based simulations, an approach have been previously proposed based on a centralized method, but this solution performs load re-distributions based on a central element, introducing global synchronization in the system. Therefore, avoiding the issues caused by centralization, a distributed, hierarchical balancing design is proposed to dynamically organize the load through three phases: monitoring, redistribution, and migration. The proposed scheme addresses improvement of fault tolerance, decrease of balancing overhead, and reduction of delays and bottlenecks, while exhibiting performance similar to the centralized approach in the experiments.