Date of Award

8-2011

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Legacy Department

Computer Science

Advisor

Martin, James J

Committee Member

Westall , James M

Committee Member

Grossman , Harold C

Committee Member

Dean , Brian C

Abstract

Modern broadband internet access cable systems follow the Data Over Cable System Interface Specification (DOCSIS) for data transfer between the individual cable modem (CM) and the Internet. The newest version of DOCSIS, version 3.0, provides an abstraction referred to as bonding groups to help manage bandwidth and to increase bandwidth to each user beyond that available within a single 6MHz. television channel. Channel bonding allows more than one channel to be used by a CM to provide a virtual channel of much greater bandwidth. This combining of channels into bonding groups, especially when channels overlap between more than one bonding group, complicates the resource allocation problem within these networks.
The goal of resource allocation in this research is twofold, to provide for fairness among users while at the same time making maximum possible utilization of the available system bandwidth. The problem of resource allocation in computer networks has been widely studied by the academic community. Past work has studied resource allocation in many network types, however application in a DOCSIS channel bonded network has not been explored.
This research begins by first developing a definition of fairness in a channel bonded system. After providing a theoretical definition of fairness we implement simulations of different scheduling disciplines and evaluate their performance against this theoretical ideal. The complexity caused by overlapped channels requires even the simplest scheduling algorithms to be modified to work correctly.
We then develop an algorithm to maximize the use of the available system bandwidth. The approach involves using competitive analysis techniques and an online algorithm to dynamically reassign flows among the available channels. Bandwidth usage and demand requests are monitored for bandwidth that is underutilized, and demand that is unsatisfied, and real time changes are made to the flow-to-channel mappings to improve the utilization of the total available bandwidth.
The contribution of this research is to provide a working definition of fairness in a channel bonded environment, the implementation of several scheduling disciplines and evaluation of their adherence to that definition, and development of an algorithm to improve overall bandwidth utilization of the system.

Share

COinS