Load balanced implementation of standard clock method

Standard clock approach is used to simulate a number of parametric variants of a single system. In this paper, we discuss a distributed implementation of the standard clock approach to simulation on networks of heterogeneous UNIX workstations. The implementation follows the client/server model, each server process that runs on a separate workstation, simulates a certain number of variants. The goal of the research is to determine the reduction in the total simulation time as more workstations are added to the run. The simulation completion time, which is considered as the performance measure, depends on the assignment of variants to workstations. We use two different load balancing techniques: (1) a static load balancing that is based on estimated cost of each variant and (2) a dynamic load balancing that migrates variants between workstations, based on their estimated performance during the simulation process. Simple queuing models are used to study the performance. Numerical results obtained from real-time simulations on a network of up to seven workstations are used to investigate the speedup and the efficiency of both the implementation and the load balancing techniques.