Research On Dynamic Load Balancing For HLA Simulation With Intra-federate Parallelism

During HLA (High Level Architecture) distributed simulation runtime, the executionperformance may be seriously affected due to the simulation load imbalance, includingboth computation load and communication load. On the other side, as the scale andcomplexity of HLA simulation increase, the intra-federate parallel simulation strategy hasbeen becoming an effective solution, to improve the efficiency of simulation. Hence, inthis paper, the load balancing problem of HLA simulation supporting the intra-federateparallel LPs (Logical Processes) is considered, and a dynamic load balancing approach isproposed to improve the execution performance.The dynamic load balancing process consists of two stages: the first is initial staticpartition of LPs, and the second is dynamic repartition of LPs during simulation runtime.In the initial stage, the LPs to be simulated are partitioned into several parts according totheir characteristics, namely, the distribution of simulation events and interactions, thus thestatic load balance can be achieved. Then each part of the LPs are allocated to a parallelfederate model, which concurrently simulates multiple LPs inside each federate. Duringsimulation runtime, the states of the both types of simulation load will be detected, and theLPs will be dynamically repartitioned when the load imbalance occurs. Furthermore, tocomplete the redistribution of LPs, an LPs migration mechanism is applied.As a result, our approach can realize the dynamic balancing of computation load andcommunication load. Moreover, the interactions between federates can be reduced asmuch as possible during the simulation process. Meanwhile, the experiment resultsindicate that a better performance can be obtained by applying the approach when runningthe HLA simulation with intra-federate parallelism.