An improved dynamic load balancing algorithm for parallel microscopic traffic simulation

Domain decomposition is a key technique for the efficient implementation of parallel microscopic traffic simulation (PMTS) in large-scale urban traffic scenarios. However, current domain decomposition methods don't take the dynamic property of traffic flow into account, so load imbalances among processors are unavoidable during the simulation run. In this paper, an improved distributed dynamic load balancing algorithm based on Receiver Initiated Diffusion scheme is proposed. In this algorithm which adopts near-neighbor information as the basis of load balancing determination, the under-loaded processor requests load from overloaded neighbors. Then the parallel USTCMTS2.0 based on high-performance computer KD50-I-E is adopted as the platform to implement and evaluate that algorithm. The experiment results indicate that our algorithm achieves better load balancing and more efficient parallel performances than the related work.