Efficient simulation of TCP/IP for mobile wireless communications using importance sampling

Future military communications systems are expected to exchange information in a seamless and reliable manner across heterogeneous networks composed of both fixed, wired nodes and mobile, wireless nodes. While commercial protocols such as TCP/IP have improved the seamlessness of military communications systems, the use of TCP/IP in the mobile, wireless environment for which it was not designed can have a deleterious effect on reliability. Reliable performance of communication systems is ultimately limited by the occurrence of rare events, such as the probability of packet loss. Here we consider a model of the behavior of TCP/IP at a wired-wireless interface in the network which requires the use of Monte Carlo (MC) simulation to estimate the rare event probabilities. To achieve the large speedup required to make MC simulation feasible, we adapt our novel importance sampling (IS) technique called direct probability redistribution (DPR) to the problem of estimating packet loss at the wired-wireless interface in TCP/IP based networks. We demonstrate that DPR is an efficient simulation technique capable of working in the feedback environment resulting from TCP/IP, and obtain orders-of-magnitude speedup of the Monte Carlo simulation used to estimate the packet loss rate in such systems.