Large Scale Simulation - Convolution of High Fidelity Model

Large scale simulation is essential for understanding the performance of mobile wireless Networks consisting of several thousand nodes with multiple types of radios, each with diverse characteristics and different types of subnet structure. A tradeoff is required between modeling fidelity on one hand, and the type of radio and number of nodes on the other, and typically, results are unsatisfactory for large scale simulations. In this paper, we present an approach to abstract large scale simulation which first synthesizes the results of limited size simulation models and then combines them using convolution techniques to project performance results spanning several subnets. This technique essentially breaks the problem domain into several independent dimensions appropriate to the radios and their subnets, such as waveforms in use W, traffic load ranges for each load L, and message priority P. Each such tuple (W,L,P) defines a performance region that can be represented by analytic means, such as a probability density function. The convolution of such functions over any number of regions traversed by a message will then yield the total delay and message completion rate characteristics of that message. The implementation of this model yields achievable real-time results that are independent of the number of nodes but dependent on the granularity of W, L, and P.