Adaptive Distributed Simulation for Computationally Intensive Modelling

The potential gain in speeding up the execution time of computationally intensive simulation in workstation clusters is tremendous. One of the challenges for exploring this gain is adaptation of the dynamic workload conditions and the heterogeneity of the workstation clusters. Static schemes of allocating simulation workload to distributed workstations are inadequate because the assigned workload may change over time and the workstation usage may vary as a result of the workload of other users. An adaptive distributed simulation should also take account of the diierences in workstation architectures and networking technologies among workstation clusters. Hence, we propose an adaptive scheme that integrates a local approach and a global approach of workload distribution to achieve automatic adaptation at runtime. In our experiments, the eeectiveness of this scheme is demonstrated by the execution proole of a distributed simulation of a broadband communication network. We observe that the simulation exhibits a good speedup throughout its runtime under diierent conditions of resource availability in the workstation clusters.