Heuristics for dynamic load balancing in parallel computing

In parallel computing, dynamic load balancing of parallel codes is considered as a crucial problem. The goal is to distribute roughly equal amounts of computational load across a number of processors, while minimizing inter-processor communication. The objective is to optimize the time of the simulation execution. In some applications, the load grow in unpredictable way that is why another distribution must be computed dynamically. Graph partitioning and repartitioning are usually combined to solve the dynamic load-balancing problem. In this paper we study and evaluate heuristic partitioning methods such as region expansion, multilevel, kernighan-lin algorithms; And methods of repartitioning graphs with a comparison between these different methods. Advantages and limitations of different existing heuristics in the literature are cleared.