Whole file
Slides for Oral Defence
Source code of the simulation program

Abstract

In data networks, there are two types of flows, elastic and inelastic. Elastic flows can work with a wide range of throughput but inelastic flows can only proceed with a particular data rate, not higher and not lower. Elastic flows is the dominating type in the Internet. There is abundant prior work in the literature on how to regulate elastic data flows in the network. The introduction of inelastic flows to coexist with elastic flows creates a problem. Inelastic flows, by definition, do not want to be subjected to congestion control; yet without control, they are likely to compete unfairly with elastic flows.

This thesis is a study of such traffic coexistence. We study several different control mechanisms that can be adopted by the inelastic flows, namely TCP-friendly congestion control, a variety of admission control and no control. We then formulate metrics and methodologies to evaluate these control mechanisms. Finally, we discuss the feasibility and concerns of implementing these control mechanisms to shed a light on the way to realise our proposals.

Our work shows that in a traffic coexistence environment, the inelastic flows can use admission control besides congestion control to make the resource allocation fair. Moreover, admission control is more suitable for inelastic flows because it makes the throughput of inelastic flows predictable. According to our analysis, certain ways of doing admission control can perform no worse than TCP-friendly congestion control in various metrics. Therefore, admission control on traffic coexistence environment is a field of worthwhile study.

Table of Contents

  • Abstract
  • Part 1: Background
      1. Background on coexistence
      1. Model of Heterogeneous Flows
  • Part 2: Evaluation
      1. Stability of network under different controls
      1. Bandwidth allocation
      1. Evaluation based on utility functions
      1. Blocking probability
      1. Population
  • Part 3: Conclusion
      1. Conclusion
  • Appendices
    • A. Glossary
    • B. Introduction to Poisson counter driven stochastic differential equations
    • C. Simulation