he College of Engineering and Computing, George Mason University
Massieh Kordi Boroujeny
Bachelor of Science,
Master of Science,
One of the main challenges in modern communication networks, like the Internet, is providing performance guarantees, such as bounds on end-to-end delay, while avoiding underutilization
of the network resources. In the early 1990s an approach to address this problem was proposed in which the input traffic was bounded either stochastically or deterministically by a so-called traffic envelope. Network calculus was developed to derive end-to-end
delay bounds from the traffic burstiness bounds. Since deterministic network calculus can lead to loose bounds, our research focuses on stochastic network calculus.
In this dissertation, we address three open problems in applying stochastic network calculus to practical networks: 1) estimating an appropriate stochastic traffic envelope
for an arbitrary traffic source; 2) enforcing conformance of a given traffic flow to a stochastic traffic envelope; 3) admission control based on an enforceable traffic envelope, while achieving statistical multiplexing gain. We develop a method to characterize
an arbitrary traffic source by a traffic envelope that takes the form of a phase-type distribution. The versatility and generality of the phase-type distribution make it useful for obtaining tight bounds to characterize the traffic. We particularize a class
of stochastic burtiness bounds using the proposed phase-type bounds. We also develop a stochastic traffic regulator that forces a traffic flow to conform to a given traffic envelope from a class of traffic envelopes, including our proposed phase-type envelopes.
We propose a new traffic envelope, referred to as the W-envelope, based on the moment generating function of the workload process obtained from offering the traffic to a constant service rate queue. We show how the W-envelope satisfies can be used in a QoS
(quality of service) framework for provisioning stochastic end-to-end delay guarantees in conjunction with the proposed traffic characterization method and stochastic traffic regulator. Finally, we develop a new available bandwidth estimation (ABE) method
that can provide accurate estimates of the available bandwidth on an end-to-end network path even in presence of packets dropped due to congestion. Our ABE method could be used to discover the amount of available bandwidth on an end-to-end, which could then
be used to provide stochastic delay guarantees for time-sensitive traffic via our proposed QoS framework.