Abstract: In parallel and distributed discrete-event simulation, the work load of logical processes (LPs) may become imbalanced due to the change of resources or newly generated events. This will increase LPs' waiting time in conservative simulations and the rollback times in the optimistic simulation. A topology-aware dynamic load-balancing approach of conservative simulation was proposed to solve this problem. The working time was monitored and calculated to detect whether imbalance occurs based on K-means clustering algorithm. The load balancing scheme was calculated according to the load and topology. Experimental results show that the proposed approach can balance the load rapidly and improve the performance of the simulation.

Key words: parallel simulation, dynamic load-balancing, cluster analysis, load-imbalance detection