Résumé: We consider the problem of channel sharing by rate adaptive streams belonging to various classes. The performance metric per class is the mean bandwidth allocated to connections in the class possibly scaled by appropriate factors. We first provide a bandwidth adaptation policy that maximizes a linear combination of class performance metrics; then, we use this result to characterize the region where the class performance metrics lie under any bandwidth adaptation policy. Based on the results above we use stochastic approximation techniques to provide a policy that optimizes a combination of concave rewards associated with class performance metrics. Finally, we propose a modification of the optimal policy to account for the case where connection holding times are unknown, and study its performance through simulations.

Langue: Anglais