Revenue Improvement Through Demand-Dependent Pricing of Network Services
Abstract
This thesis addresses the issue of efficient resource allocation in broadband networks through the pricing of guaranteed services. Resource allocation in networks is typically achieved through technical methods such as scheduling, routing, and admission control, however these techniques are not predictive in terms of the expectation of rewards based upon variable demands. This work shows that revenue improvement can occur in this network environment when a dynamic pricing policy is applied as opposed to optimal static pricing. The network model used is that of a broadband channel with multiple classes of service, each of which has a Poisson arrival process with exponential service times, where the arrival rate is a function of the current price and a partially observed Markov chain. The optimization model is a continuous time, average-reward, partially observed Markov decision process, with the state defined as the profile of current active sessions (of various types) and the probability distributions over the demand states for each service class. We have developed a heuristic that seeks improvement of the objective based upon the probability that the service class is in one of several possible demand states. Our heuristic involves a state estimation procedure to determine: (1) when the system has transitioned to a new demand state, and (2) when and what control (price) should be applied to improve revenue. This work contributes to the large body of recent research into network pricing which primarily examines the concept from a static point of view. Most analysis in this area of research assume a constant load upon the system and does not explicitly accommodate the stochastic nature of arrivals and departures, as does our heuristic. In the more common network pricing formulation, demand is assumed to be a constant and known function of price, and prices are adjusted to find an optimal mix of the user classes.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 2000
- Accession Number
- ADA377505
Entities
People
- David M. Sanders
Organizations
- University of Virginia