Research on Dynamic Load Balancing Algorithms for Parallel Transportation Simulations

Dongliang Zhang, Changjun Jiang, Shu Li

Abstract: To the issue of dynamic load balancing in parallel transportation simulations, we describe two algorithms for different types of task partitions, parallel lines partition and grid partition. In the algorithms, load balance is obtained by iteratively moving the boundary lines according to the relative balance of adjacent sub-domains. Assuming real traffic distribution as the experimental work load, we test the performance of the algorithms. And the result we observe confirms the value of the methods. Based on the discussion of the communication overheads under different types of partitions, due to the relative small amount of boundary lines, grid partition can decrease the communication overheads and is a more adaptive partition model.