It is well known that the performance of a parallel discreteevent simulation (PDES) depends on the allocation of the workload to processors. In particular, poor performance may be the result of an imbalance of the workload on the processors. In earlier work, we examined automated load balancing techniques that statically allocated the workload based on prior run-time data. However, a good initial distribution of the workload may result in a load imbalance if the characteristics of the simulation change over time. Furthermore, some of the processors may gain external workloads at some point in time. Thus, there is a need for dynamic load balancing methods that can adapt to changes in the simulation or the system. In this paper, we discuss our experiments in dynamic load (object) migration and load balancing using the SPEEDES simulation framework.
Linda F. Wilson, Wei Shen