Abstract:

The platforms equipped with multi-core central processing units (CPU) and graphics processing units (GPU) provide a new hardware for executing parallel agent-based simulations (PABS) and the load balancing mechanism is an effective approach to make the best of hardware resources and improve the running performance of parallel simulations. In order to solve the load balancing for PABS on multi-core CPU and GPU heterogeneous platforms, a multilayer load distributing model for PABS is built and a static load partitioning algorithm based on the constrained k-means clustering algorithm and a dynamic load balancing algorithm are proposed. The method of finding interactions between partitions is presented so as to avoid a lot of unnecessary computations between agents that actually can not interact with each other. The experiments show that the mechanism is viable and effective.