Anticipatory routing for highly mobile endpoints

We consider the problem of routing to endpoints with very high "effective" mobility, i.e, when the period between changes in an endpoint's location is comparable to the time it takes for the location tracking mechanism to converge. This could happen due to increased endpoint speed, increased control message latency, or decreased cell size. When this happens, conventional location tracking approaches fail - by the time such mechanisms converge, the endpoint has already moved to a new location, and we say the "reactive limit" has been reached. When mobility exceeds the reactive limit, prediction techniques are required for network control. We describe "anticipatory routing", a mechanism that uses a limited history of past movement of the endpoint to predict its future locations. In particular, we use linear regression to predict future locations and affiliation/departure times of the endpoint. Using a simulation model of a multi-hop wireless network, we study the performance of anticipatory routing. Our results indicate that above a certain mobility threshold, anticipatory routing significantly outperforms conventional location tracking mechanisms. Its magnitude of improvement increases with higher mobility - in particular, its rate of throughput degradation with speed is 56% better.