Optimizing revenue for bandwidth auctions over networks with time reservations

This paper concerns the problem of allocating network capacity through periodic auctions, in which users submit bids for fixed amounts of end-to-end service. We seek a distributed allocation policy over a general network topology that optimizes revenue for the operator, under the provision that resources allocated in a given auction are reserved for the entire duration of the connection. We first study periodic auctions under reservations for a single resource, modeling the optimal revenue problem as a Markov Decision Process (MDP), and developing a receding horizon approximation to its solution. Next, we consider the distributed allocation of a single auction over a general network, writing it as an integer program and studying its convex relaxation; techniques of proximal optimization are applied to obtain a convergent algorithm. Combining the two approaches we formulate a receding horizon optimization of revenue over a general network topology, leading to a convex program with a distributed solution. The solution is also generalized to the multipath case, where many routes are available for each end-to-end service. A simulation framework is implemented to illustrate the performance of the proposal, and representative examples are shown.