The University of British Columbia

We present the Service Directory Placement Algorithm (SDPA), a directory-placement scheme that leverages the performance of existing service discovery protocols over wireless ad-hoc networks. SDPA promotes the deployment of a nomadic service directory, whose current location in the network varies according to the dynamics of service-discovery queries driven by the users' applications and partial knowledge of the network's topology. SDPA is based on a heuristic approach, whose performance is optimized by formulating the directory-placement problem as a Semi-Markov Decision Process solved by means of a reinforcement-learning technique known as Q-Learning. Performance evaluations obtained through computer simulations of networks with up to 45 hosts moving at pedestrian walking speeds equal to or slower than 2 m/s reveal average bandwidth savings close to 50 #x025; over a default broadcast approach for service discovery once an efficient directory-placement policy is found.