Optimal throughput and delay in delay-tolerant networks with ballistic mobility

This work studies delay and throughput achievable in delay-tolerant networks with ballistic mobility -- informally, when the average distance a node travels before changing direction does not become vanishingly small as the number of nodes in the deployment area grows. Ballistic mobility is a simple condition satisfied by a large number of well-studied mobility models, including the i.i.d. model, the random waypoint model, the uniform mobility model and Levy walks with exponent less than 1. Our contribution is twofold. First, we show that, under some very mild and natural hypotheses satisfied by all models in the literature, ballistic mobility is strictly necessary to achieve simultaneously, as the number of nodes grows, a) per-node throughput that does not become vanishingly small and b) communication delay that does not become infinitely large. Any network whose nodes exhibit a more "local" mobility pattern (e.g. Levy walks with exponent greater than 1, or Brownian motion) must sacrifice either a) or b), regardless of the communication scheme adopted -- even with network coding. Second, we present a novel packet routing scheme. Our scheme is relatively simple and does not rely on centralized control, replication, or static base stations. At the same time it achieves both non-vanishing throughput and bounded delay as the number of nodes grows, on any network with ballistic mobility (i.e. whenever they can be simultaneously achieved), asymptotically outperforming any existing communication scheme that exploits node mobility to boost throughput.