The networking research community has recently become intrigued with the possibility of offer-ing differentiated service types which are dynamically priced to enc~6rage favorable consumption of network resources. It is widely speculated but far from clear that pricing can serve as a means of congestion control (packet regulation) as well as a means of cost recovery for infrastructure providers. This work will explore the former assertion. Current mechanisms for the control of packet flows, while often justified by hard engineering anal-ysis, are ad hoc in the sense that they are not conceived as solutions to stochastic optimal control problems. Learning is based mechanisms may be useful in coping with and differentiated services.
The research objectives of this proposal are twofold:
1. to develop technological mechanisms for packet regulation based on stochastic optimal control and to establish the general applicability of this perspective in networking through mathe-matical and computational analysis, and
2. to mitigate the effect of computational complexity in solving real-world stochastic optimal control problems and to promote the development of practical learning based computational methods.
The education objectives of this proposal are
1. to provide a vehicle by which undergraduate and graduate students will learn general approaches to research: modeling, optimization, analysis, and design; and specific technological expertise: stochastic optimal control, computational methods, and the cutting edge of data communications, and
2. to disseminate the results of the proposed research to as broad an audience as possible, including the telecommunications industry. ***
