Saturation Rate in 802.11 Revisited

There have been extensive studies on the performance of IEEE 802.11 networks, most of which focus on the saturation regime where user queues are always full. This paper departs from prior work in that we focus on non-saturated regimes. Specifically, we consider two scenarios: one consisting of a mixture of saturated and non-saturated users, the other consisting of users of buffer size one. In both cases we show that the system throughput achieved can be significantly greater than the saturation throughput achieved in a homogeneous environment of the same number of users with infinite queues. With these results we argue that the notion of saturation throughput is an inherently pessimistic one. When there is natural heterogeneity in the system or when queues are forced to be empty from time to time due to limited buffer size, the effect of statistical multiplexing emerges, which leads to increased system throughput.