Predictive Dynamic Load Balancing for Large-Scale HLA-based Simulations

Due to the dependency on the resources, HLA-based simulations can experience load imbalances and consequently loose execution performance. Such imbalances are originated from external background load, inappropriate deployment of simulation entities, heterogeneity of resources, and dynamic load variations. The High Level Architecture (HLA) was developed aiming to facilitate the creation and control of distributed simulations through a design framework and management services, but such an architecture does not offer solutions for solving load imbalance issues. In order to provide mechanisms for preventing performance loss caused by load imbalances in distributed simulations, numerous balancing approaches have been developed. The majority of these mechanisms present application-specific solutions or limited awareness of environment characteristics. To cope with this problem, a distributed dynamic balancing scheme has been designed, but its redistribution algorithm, as other developed balancing schemes, is limited to just correct load distribution issues and does not react properly in presence of abrupt load changes due to be based on recent load status. Therefore, a predictive balancing scheme is proposed to provide a method to decrease the number of precipitated migration moves and to detect and prevent load imbalances based on load variation tendencies. In order to observe and evaluate the proposed scheme, experiments have been performed to compare performance gain and efficiency with the distributed balancing scheme.