During the past years, there has been an increasing interest in the design and development of network traffic controllers capable of ensuring the QoS requirements of a wide range of applications. In this thesis, based on previous work, we use a systems approach to construct a dynamic model for the token bucket mechanism: a traffic controller widely used in various QoS-aware protocol architectures. An enhanced model of multiplexor is also added into the multiple token buckets system model. In this way, the model represents a complete system at the access node of the network. We then develop an optimization algorithm based on a dynamic programming and genetic algorithm approach. Applying two MPEG-1 video traces and two self-similar traffic traces, we conduct an extensive campaign of numerical experiments allowing us to gain insight into the operation of the controller and evaluate the benefits of using a genetic algorithm approach to speed up the computation process based on dynamic programming. Our results show that the optimization is not only capable of getting the best cost, but also balancing the costs corresponding to different aspects. The results also shows that the use of the genetic algorithm proves particular useful in reducing the computation time required to optimize the operation of a system consisting of multiple token-bucket regulated sources.

In this thesis, we construct traffic models, which exhibit both short-range dependence (SRD) and long-range dependence (LRD) using Poisson process and doubly stochastic Poisson (or Cox) process (DSPP) driven by fractional Brownian motion (FBM). We also develop a novel dynamic system model for the token bucket (TB) control algorithm used in computer networks. In this model, token buckets police incoming traffic; and one multiplexor serving all the token pools multiplexes conforming traffic using round robin scheme. The state of the system is formally defined and control strategies are also proposed. We study several issues related to performance corresponding to different stochastic inputs using the proposed model located at the edge of the backbone network. The numerical results demonstrate that this system can be adapted to any kind of stochastic traffic. The results can be served as a tool for the designers of such controllers to set up different system parameters.

The main objective of the different security services and mechanisms today are to provide privacy of information and to ensure that the tools used to establish a proper environment to the user are reliable and trusted. To provide a high level of security, an exhaustive approach that depicts a complete package of the Internet security classification categories is needed to show how, where and when each of the security services works. The proposed comprehensive Internet security taxonomy is innovative in its overall study of Internet security as the guidelines established to assess the taxonomy are strict. Many of the known Internet security attacks and the Internet security services associated with them are analyzed in this proposed taxonomy. A mapping of the security services against the security attacks and corresponding countermeasures is given. An assessment of the performance of the proposed taxonomy is also given, showing it to be useful, exhaustive and unambiguous.

Operating in a more constrained and more demanding system, DFO has been required to revise and revitalize their strategic plan that brings the requirement of a more transparent, participatory, and informed scientific decision making process through sound science. As for scientific organizations as DFO science, Analytic Hierarchy Process that provides a multiple criteria scoring result based on participants' knowledge and expertise is considered for the science prioritization supporting the decision making. For the specific situation in DFO, considering the organizational arrangement and the delivery of science function, different methods have been formulated and simulated to prove that AHP is an appropriate approach. Meanwhile, to perform the risk analysis for DFO science prioritization, expected utility theory has been chosen as the proper approach to take risk and uncertainty into account. Application software will be developed using these approaches specifically for DFO science risk-based prioritization process and an experiment carried out in Maritimes BIO workshop will be used to demonstrate the approaches and application system proposed in this thesis.

Quality of Service (QOS) of a congested network can be evaluated by some key factors such as packets discarding rate, queueing delay and the throughput of a bottleneck buffer. A network designer needs to predict these parameters in order to dimension a network properly. This thesis developed a solver for predicting the stability of a TCP network when servers implement Active Queue Management (AQM) schemes. The solver analyzes two AQM schemes. One is Random Early Detection (RED) and the other is Dynamic Random Early Detection (DRED). We verify the analysis of the solver through the simulation of a small network with multiple bottlenecks carrying a large number of TCP connections.

This thesis introduces new approaches, namely the DataBoost and DataBoost-IM algorithms, to extend Boosting algorithms' predictive performance. The DataBoost algorithm is designed to assist Boosting algorithms to avoid over-emphasizing hard examples. In the DataBoost algorithm, new synthetic data with bias information towards hard examples are added to the original training set when training the component classifiers. The DataBoost approach was evaluated against ten data sets, using both decision trees and neural networks as base classifiers. The experiments show promising results, in terms of overall accuracy when compared to a standard benchmarking Boosting algorithm. The DataBoost-IM algorithm is developed to learn from two-class imbalanced data sets. In the DataBoost-IM approach, the class frequencies and the total weights against different classes within the ensemble's training set are rebalanced by adding new synthetic data. The DataBoost-IM method was evaluated, in terms of the F-measures, G-mean and overall accuracy, against seventeen highly and moderately imbalanced data sets using decision trees as base classifiers. (Abstract shortened by UMI.)

In this thesis, dynamic mathematical models are constructed to describe the population distribution in Canada based on the model in previous work by Ahmed and Rahim [1]. Numerical results demonstrate that the model population is in close agreement with the actual population. This indicates that the presented model can be used as a valuable tool for describing the dynamics of population distribution. We also demonstrate that by using modern Systems and Optimal Control theory [2], it is possible to formulate optimum immigration and job creation strategies while maintaining population level close to certain pre-specified targets. An optimization algorithm [2] is then developed based on dynamic programming and gradient algorithm approach. Unknown parameters such as birth rate, death rates and transition rates are estimated and identified. The system model obtained by using the identified parameters is then augmented by adding a fourth equation describing the dynamics of unemployment rate. This model is then used to formulate a control problem with immigration and job creation rates being the decision (control) variables. Using optimal control theory, optimum immigration and job creation policies are determined. Results are illustrated by numerical simulation and they are found to be very encouraging.

Infinite variance distributions are among the competing models used to explain the non-normality of stock price changes (Mandelbrot, 1963; Fama, 1965; Mandelbrot and Taylor, 1967; Rachev and Samorodnitsky, 1993). We investigate the asymptotic option price formula in infinite variance setting for both independent and correlated data using point processes. As we shall see the application of point process models can also lead us to investigate a more general option price formula. We also apply a recursion technique to quantify various characteristics of the resulting formulas. It shows that such formulas, and even their approximations, may be difficult to apply in practice. A nonparametric bootstrap method is proposed as one alternative approach and its asymptotic consistency is established under a resampling scheme of m = o(n). Some empirical evidence is provided showing the method works in principle, although large sample sizes appear to be needed for accuracy. This method is also illustrated using publicly available financial data.

A Verilog HDL digital circuit fault simulator to detect permanent stuck-at logic faults for embedded cores-based synchronous sequential circuits is proposed in this thesis. The fault simulator can emulate a typical built-in self-testing (BIST) environment that utilizes a test pattern generator that sends its outputs to a module under test (MUT), with the resulting output from the MUT being fed into a test data analyzer. A fault is detected if the module response is different from that of the fault-free MUT. The fault simulator is suitable for testing synchronous sequential circuits described at the gate and flip-flop level in Verilog HDL. The subject thesis describes the detailed architecture and implementation of the fault simulator. Some simulation experiments on ISCAS 89 sequential benchmark circuits are also provided and discussed. The thesis also explores possible application of the ideas proposed to current embedded cores-based systems-on-chip (SOC) technologies, specifically in the context of testing memory-based synchronous digital systems.

Implementation of fault testing environment for embeded cores-based digital circuits is a challenging endeavor. The subject thesis aims developing techniques in design verification and test architecture utilizing well-known concepts of hardware and software co-design. There are available methods to ensure correct functionality, in both hardware and software, for embeded cores-based systems but one of the most used and acceptable approaches to realize this is through the use of design for testability. Specifically, applications of built-in self-test (BIST) methodology in testing embeded cores are considered in the thesis, with specific implementations being targeted towards ISCAS 85 combinational benchmark circuits. Experimental results provided in the thesis prove the validity and importance of the approaches proposed for the design verification and test based on hardware and software co-design concepts utilizing Altera MAX Plus II simulation environment.