Modeling of long-term dependent VBR video sources and implementation of Web-based network management
Abstract
When modeling network traffic, time series are often assumed to be uncorrelated or "short-term" correlated, meaning that their autocorrelation function decays exponentially fast. A number of recent studies have shown, however, that both network traffic measurements and VBR video frame sizes are "long-term" dependent. That is, the autocorrelation function decays only polynomially fast. For such time series, there is no natural length of "burst"; bursts appear on a wide range of time scales. In this study, we (1) show that the Hurst parameter, commonly used to characterize long-term dependent processes, is not a good indicator of queue performance with VBR video sequences as input; (2) derive a new parameter called the Fluctuation Energy, and show that it provides a good characterization of VBR video sequences; (3) propose a new VBR video source model based on the Fluctuation Energy; (4) derive the queue performance with input generated by the new model using a "queues in a random environment" model. We have also developed an integrated system that uses the technologies of the World Wide Web to assist in the management of complex ATM and HiPPI networks. The system provides all the management functions of FORE Systems' ATM Network Interface (AMI) and Essential Communications' HiPPI Management Interface. Moreover, it can control the generation of video streams on the network, capture traffic measurements at any ATM endpoint, and generates useful statistics on the fly. A network resource discovery and sharing framework is also studied as an extension of the above system.
Degree
Ph.D.
Advisors
Coyle, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.