Title:
Pricing Network Resources for Differentiated Service Networks
Pricing Network Resources for Differentiated Service Networks
Author(s)
Yang, Weilai
Advisor(s)
Blough, Douglas M.
Owen, Henry L., III
Owen, Henry L., III
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
We developed a price-based resource allocation scheme for Differentiated Service (DiffServ) data networks. The DiffServ framework was proposed to provide multiple QoS classes over IP networks. Since the provider supports multiple service classes, we need a differentiated pricing scheme, as supposed to the flat-rate scheme employed by the Internet service providers of today. Charging efficiently is a big issue. The utility of a client correlates to the amount of bandwidth allocated. One difficulty we face is that determining the appropriate amount of bandwidth to provision and allocate is problematic due to different time scales, multiple QoS classes and the unpredictable nature of users.
To approach this problem, we designed a pricing strategy for Admission Control and bandwidth assignment. Despite the variety of existing pricing strategies, the common theme is that the appropriate pricing policy rewards users for behaving in ways to improve the overall utilization and performance of the network. Among existing schemes, we chose auction because it is scalable, and efficiently and fairly shares resources. Our pricing model takes the system's availability and each customer's requirements as inputs and outputs the set of clients who are admitted into the network and their allocated resource. Each client proposes a desired bandwidth and a price that they are willing to pay for it. The service provider collects this information and produces parameters for each class of service they provide. This information is used to decide which customers to admit.
We proposed an optimal solution to the problem of maximizing the provider's revenue for the special case where there is only one bottleneck link in the network. Then for the generalized network, we resort to a simple but effective heuristic method. We validate both the optimal solution and the heuristic algorithm with simulations driven by a real traffic scenario. Finally, we allow customers to bid on the duration for which the service is needed. Then we study the performance of those heuristic algorithms in this new setting and propose possible improvements.
Sponsor
Date Issued
2004-04-12
Extent
1702010 bytes
Resource Type
Text
Resource Subtype
Dissertation