Measuring and Analyzing Migration Delay for the Computational Load Balancing of Distributed Virtual Simulations

摘要

Distributed virtual simulations can always undergo load imbalances during run-time due to their dependency on underlying shared resources. Such imbalances are commonly generated by external background load, improper deployment of simulation elements, and dynamic oscillations of simulation load. HLA was devised as a simulation framework that simplifies the design and management of distributed simulations. Even though the framework presents services for the coordination of simulations, it does not provide any tool for identifying and reacting to load imbalances. Due to the importance of balancing simulation load, many schemes have been developed, aiming to reduce simulation time through analysis of specific metrics. These schemes are limited to issues from specific simulation applications, or they disregard large-scale environmental characteristics. In order to overcome the drawbacks of previous balancing approaches, a distributed balancing scheme has been designed. Nevertheless, this scheme, as well as the others, is not concerned with migration latencies when redistributing simulation load. Migration delays are directly involved with balancing responsiveness and efficiency, and they can cause simulation performance loss instead of performance improvement if they are not considered in the redistribution analysis. Thus, a migration-aware balancing scheme is proposed to include migration latency analysis in its load redistribution algorithm. Extensions are also proposed to improve the analysis of migration delays by enabling necessary costly migrations in iterative analysis. Experiments have been conducted to evaluate the proposed migration-aware balancing schemes by comparing performance when costly migrations are present in simulations.