Distributed resource management and matching in sensor networks

We consider a scenario in which there are resources at or near nodes of a network, which are either static (e.g. fire stations, parking spots) or mobile (e.g. police cars). Over time, events (fires, crime reports, cars looking for parking) arise one-by-one at arbitrary nodes, and need to be quickly matched to and serviced by an appropriate nearby resource, without knowledge of future requests, and without the ability to alter any decision once it has been made. We develop distributed algorithms to direct available resources in the network to these events (or vice versa) in a coordinated fashion, so that no two resources are assigned to the same event, and the total distance of the events from their matched resources is minimized. The key idea is to extract, in a preprocessing stage, a well-separated tree metric that approximates the original network metric by a logarithmic distortion, allowing greedy matching algorithms to generate close to optimal matchings, and enabling communication-efficient probing-based algorithms for events to detect nearby available resources. The distributed matching algorithm requires no global coordination and achieves polylogarithmic performance ratio in both online and offline settings. Simulation experiments corroborate the theoretical results on solution quality and further evaluate the communication costs of our scheme in practice.