Date of Award

Spring 2010

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)

Department

Electrical and Computer Engineering

First Advisor

Roberto Rojas-Cessa

Second Advisor

Nirwan Ansari

Third Advisor

Sotirios Ziavras

Fourth Advisor

Yanchao Zhang

Fifth Advisor

George Lapiotis

Abstract

The emerging network traffic with various Quality-of-Service (QoS) requirements creates a demand for QoS service provisioning beyond the best effort service that Internet currently provides. QoS provisioning requires a framework that satisfies users' QoS and cost demand while maximizes benefits for network service providers. It is considered that QoS provisioning involves three issues: a) estimations of the network QoS performance, which can be achieved by performing network measurement; b) dissemination of the measured QoS states throughout the network with states exchanged among different network routers; and c) QoS routing. In this dissertation, these three issues are addressed.

In QoS networks, knowledge of the network status is crucial but current implemented network protocols cannot provide enough QoS measurement functions. For such purposes, a network measurement framework is proposed that runs active measurement tools to estimate multiple QoS classes. An important issue involved in the network measurement is the tasks conflict problem. This problem occurs when multiple active measurement tasks sending probing packets in the same network segment at the same time, and causes misleading report of QoS performance because the tasks' contention for network resources disturb each other's measurement. In this dissertation a novel scheduling algorithm is proposed to allow such contention among measurement tasks and to shorten the measurement period time.

In addition, flooding algorithm is dominantly used in link state dissemination. It faces the large overhead problem when used in QoS networks since a lot of QoS states need to be updated frequently. On the contrary, the alternative algorithm, single spanning tree dissemination, may not be able to achieve fast convergence or reliability. In the proposed new scheme, Per-Hop pArtial-Spanning Tree Adjust (PASTA) for dissemination, link states can be distributed with low overhead and fast speed, and the computation complexity to build the tree is small compared to single spanning tree algorithm. The reliability of the dissemination is enhanced by the multi-spanning-tree approach and the back-trace method.

Furthermore, in most of the current QoS architectures, a traffic flow receive the unified service at every hop on the path, resulting that the end-to-end QoS provisioning lacks flexibility and granularity. To solve this problem, a nested DiffSery model is presented. In this model, service at each hop is quantified and divided into multiple classes, and the edge router on behalf of user is allowed to select different service at each hop. Under this framework, routing in terms of cost and QoS requirements can be regarded as a delay-leastl-ow-cost (DCLC) problem, which is known to be NP-hard. An improved k-shortest path QoS routing algorithm is proposed to solve this problem.

Share

COinS