Abstract

This thesis presents a new approach for dynamic optimization of real-time distributed multi-body simulations. It continuously optimizes network communication for a given model partitioning. The optimization problem proposed is to minimize a network performance cost function using the recently developed dynamic simplex method (DSM). This iterative minimization approach requires a small number of function evaluations, no derivatives, and it can track a moving optimum in a stable manner. Furthermore, evidence indicates that the approach converges close to the optimal solution provided it doesn't vary too rapidly. Experimental investigations are performed on a time division multiple access (TDMA) computational network to validate the new approach.