On-line tuning of prices for network services

Recent investigations into the pricing of multiclass loss networks have shown that static prices are optimal in the asymptotic regime of many small sources. These results suggest that nearly optimal prices for highly aggregated systems can be computed from the solution to a limiting deterministic optimization model. When the assumption of many small sources does not hold, static prices are still preferable (for practical reasons), but we are left with the difficult issue of computing an optimal solution when the stochastic nature of the process cannot be ignored. In this paper, we develop a computational procedure for optimizing static prices that operates by adjusting prices in response to actual customer arrivals and departures and is robust to parametric uncertainty about the underlying system. We provide initial arguments for the convergence properties of our optimization algorithm, and we illustrate its application in several numerical examples.