Search

The recent exponential growth of Information and Communication Technologies (ICT) also leads to a rapid increase in ICT energy consumption. The challenge of achieving energy-efficient ICT has been therefore explored, including Energy Harvesting (EH) technologies with intermittent renewable energy sources. Thus there has been considerable interest in understanding how to optimise energy efficiency and quality of service (QoS) for computer-communication systems powered by EH technologies. This thesis investigates mathematical modelling and performance evaluation for computer communication systems powered by EH technologies with intermittent renewable energy sources. In our research, we use the Energy Packet Network (EPN) paradigm, which is a discrete state-space modelling framework based on G -network theory. In such a system, discrete activities such as computer jobs and data in the form of packets consume energy represented by discrete energy packets (EPs), each of which is a basic unit of energy in Joules. This approach uses queueing theory such that the joint behaviour of discretised energy flows and job or data flows is analysed in a single model, which can evaluate both the performance and energy efficiency of a complex interconnected computer-communication system. Specifically, our work uses the EPN paradigm to address four relevant problems of practical interest to optimise performance and energy efficiency. We use energy flow only or job flow and energy flow together to optimise a multi-server system's performance or QoS, for instance, the average response time of jobs. In the first problem (Problem P1), we investigate how to select the optimal fraction of power that is shared between heterogeneous servers, so as to minimise the average response time of jobs. We use the Lagrange multiplier method to solve Problem P1 analytically, which is subject to power constraints. We also obtain a physically meaningful condition to guarantee system stability and global optimality. The second problem (Problem P2) is to minimise the average response time of jobs by dynamically deciding whether to move jobs between servers, so as to balance the workload at each server. Problem P2 is solved numerically through the gradient descent. Regarding the third problem (Problem P3), we consider a cost function that combines both the average response time of jobs and the rate of energy loss. Then we select the optimal fraction of power shared between heterogeneous servers again to minimise the cost function. The optimal solution is obtained by solving a system of equations simultaneously. The fourth problem (Problem P4) investigates how to match the energy flow into energy buffers and job flow into servers or workstations, so as to minimise the average response time of jobs. We have given a necessary condition to match energy flow and job flow, which finds the local minimum of the average response time of jobs analytically. Throughout this thesis, we generalise the assumption of the previous work of the EPN paradigm that one EP can only be used to execute one single job. We assume that the number of jobs can be executed by one EP is a random variable of Xi following general probability distribution. The derivation of the average number of jobs that can be executed by one EP is given in Appendix A. For the sake of computational convenience, we assume the random variable X_i follows a geometric distribution in our analysis. However, we have also discussed how to solve the problem with the general distribution in Chapter 6 and Appendix B.

Software security and reliability are aspects of major concern for software development enterprises that wish to deliver dependable software to their customers. Several static analysis-based approaches for facilitating the development of secure and reliable software have been proposed over the years. The purpose of the present thesis is to investigate these approaches and to extend their state of the art by addressing existing open issues that have not been sufficiently addressed yet. To this end, an empirical study was initially conducted with the purpose to investigate the ability of software metrics (e.g., complexity metrics) to discriminate between different types of vulnerabilities, and to examine whether potential interdependencies exist between different vulnerability types. The results of the analysis revealed that software metrics can be used only as weak indicators of specific security issues, while important interdependencies may exist between different types of vulnerabilities. The study also verified the capacity of software metrics (including previously uninvestigated metrics) to indicate the existence of vulnerabilities in general. Subsequently, a hierarchical security assessment model able to quantify the internal security level of software products, based on static analysis alerts and software metrics is proposed. The model is practical, since it is fully-automated and operationalized in the form of individual tools, while it is also sufficiently reliable since it was built based on data and well-accepted sources of information. An extensive evaluation of the model on a large volume of empirical data revealed that it is able to reliably assess software security both at product- and at class-level of granularity, with sufficient discretion power, while it may be also used for vulnerability prediction. The experimental results also provide further support regarding the ability of static analysis alerts and software metrics to indicate the existence of software vulnerabilities. Finally, a mathematical model for calculating the optimum checkpoint interval, i.e., the checkpoint interval that minimizes the execution time of software programs that adopt the application-level checkpoint and restart (ALCR) mechanism was proposed. The optimum checkpoint interval was found to depend on the failure rate of the application, the execution cost for establishing a checkpoint, and the execution cost for restarting a program after failure. Emphasis was given on programs with loops, while the results were illustrated through several numerical examples.

Autonomic communication, which is the development of self-configuring, self-adapting, self-optimising and self-healing communication systems, has gained much attention in the network research community. This can be explained by the increasing demand for more sophisticated networking technologies with physical realities that possess computation capabilities and can operate successfully with minimum human intervention. Such systems are driving innovative applications and services that improve the quality of life of citizens both socially and economically. Furthermore, autonomic communication, because of its decentralised approach to communication, is also being explored by the research community as an alternative to centralised control infrastructures for efficient management of large networks. This thesis studies one of the successful contributions in the autonomic communication research, the Cognitive Packet Network (CPN). CPN is a highly scalable adaptive routing protocol that allows for decentralised control in communication. Consequently, CPN has achieved significant successes, and because of the direction of research, we expect it to continue to find relevance. To investigate this hypothesis, we research new applications and contexts for CPN. This thesis first studies Information-Centric Networking (ICN), a future Internet architecture proposal. ICN adopts a data-centric approach such that contents are directly addressable at the network level and in-network caching is easily supported. An optimal caching strategy for an information-centric network is first analysed, and approximate solutions are developed and evaluated. Furthermore, a CPN inspired forwarding strategy for directing requests in such a way that exploits the in-network caching capability of ICN is proposed. The proposed strategy is evaluated via discrete event simulations and shown to be more effective in its search for local cache hits compared to the conventional methods. Finally, CPN is proposed to implement the routing system of an Emergency Cyber-Physical System for guiding evacuees in confined spaces in emergency situations. By exploiting CPN's QoS capabilities, different paths are assigned to evacuees based on their ongoing health conditions using well-defined path metrics. The proposed system is evaluated via discrete-event simulations and shown to improve survival chances compared to a static system that treats evacuees in the same way.

Autonomous self-organising systems have gained much attention to manage complex tasks without any human interaction. Similarly, simpler systems are also required for the applications of Internet of Things, such as smart home services, wearables, smart cities and connected health systems. These simpler systems provide autonomous standalone devices for remote sensing, processing and transmission of information. However, such devices may not have a reliable connections to power mains or may not be convenient for regular battery replacements. Therefore, local energy capturing from ambient intermittent sources such as vibration, electromagnetic waves, heat or light through harvesting could be of great interest for such devices. This thesis researches mathematical modelling and performance analysis of such autonomous digital devices operating with energy harvesting from intermittent sources. The approach used in this research is based on the Energy Packet Network paradigm where the arrivals of data and energy at devices are considered as discrete random processes. The devices operate by consuming harvested energy in a discrete manner in order to process, store and transmit data (wired or wireless) in a negligible time interval, such that the operation or the workload of the devices is also modelled as a discrete random process. Probability models based on random walks and Markov chains are investigated in this study to predict effective rates at which such devices operate well for different energy consumption scenarios, and to obtain closed-form formulas for stationary probability distributions and to make further analysis on the other quantities of interest. Consequently, a "product form solution" of a cascade network of N nodes where state transitions involve simultaneous state changes in multiple nodes, due to data packets that flow through several nodes consuming energy packets is proposed. A modelling approach to evaluate the effect of several battery attacks on such devices is studied. Finally, optimum placement of wireless nodes into a region where there is a spatial continuous distribution of energy and data traffic is presented for different transmission schemes, and optimisation objectives.

The random neural network (RNN) is a mathematical model for an 'integrate and fire' spiking network that closely resembles the stochastic behaviour of neurons in mammalian brains. Since its proposal in 1989, there have been numerous investigations into the RNN's applications and learning algorithms. Deep learning (DL) has achieved great success in machine learning, but there has been no research into the properties of the RNN for DL to combine their power. This thesis intends to bridge the gap between RNNs and DL, in order to provide powerful DL tools that are faster, and that can potentially be used with less energy expenditure than existing methods. Based on the RNN function approximator proposed by Gelenbe in 1999, the approximation capability of the RNN is investigated and an efficient classifier is developed. By combining the RNN, DL and non-negative matrix factorisation, new shallow and multi-layer non-negative autoencoders are developed. The autoencoders are tested on typical image datasets and real-world datasets from different domains, and the test results yield the desired high learning accuracy. The concept of dense nuclei/clusters is examined, using RNN theory as a basis. In dense nuclei, neurons may interconnect via soma-to-soma interactions and conventional synaptic connections. A mathematical model of the dense nuclei is proposed and the transfer function can be deduced. A multi-layer architecture of the dense nuclei is constructed for DL, whose value is demonstrated by experiments on multi-channel datasets and server-state classification in cloud servers. A theoretical study into the multi-layer architecture of the standard RNN (MLRNN) for DL is presented. Based on the layer-output analyses, the MLRNN is shown to be a universal function approximator. The effects of the layer number on the learning capability and high-level representation extraction are analysed. A hypothesis for transforming the DL problem into a moment-learning problem is also presented. The power of the standard RNN for DL is investigated. The ability of the RNN with only positive parameters to conduct image convolution operations is demonstrated. The MLRNN equipped with the developed training algorithm achieves comparable or better classification at a lower computation cost than conventional DL methods.

Communication services such as telephony, broadband and TV are increasingly migrating into Internet Protocol(IP) based networks because of the consolidation of telephone and data networks. Meanwhile, the increasingly wide application of Cloud Computing enables the accommodation of tens of thousands of applications from the general public or enterprise users which make use of Cloud services on-demand through IP networks such as the Internet. Real-Time services over IP (RTIP) have also been increasingly significant due to the convergence of network services, and the real-time needs of the Internet of Things (IoT) will strengthen this trend. Such Real-Time applications have strict Quality of Service (QoS) constraints, posing a major challenge for IP networks. The Cognitive Packet Network (CPN) has been designed as a QoS-driven protocol that addresses user-oriented QoS demands by adaptively routing packets based on online sensing and measurement. Thus in this thesis we first describe our design for a novel ''Real-Time (RT) traffic over CPN'' protocol which uses QoS goals that match the needs of voice packet delivery in the presence of other background traffic under varied traffic conditions; we present its experimental evaluation via measurements of key QoS metrics such as packet delay, delay variation (jitter) and packet loss ratio. Pursuing our investigation of packet routing in the Internet, we then propose a novel Big Data and Machine Learning approach for real-time Internet scale Route Optimisation based on Quality-of-Service using an overlay network, and evaluate is performance. Based on the collection of data sampled each $2$ minutes over a large number of source-destinations pairs, we observe that intercontinental Internet Protocol (IP) paths are far from optimal with respect to metrics such as end-to-end round-trip delay. On the other hand, our machine learning based overlay network routing scheme exploits large scale data collected from communicating node pairs to select overlay paths, while it uses IP between neighbouring overlay nodes. We report measurements over a week long experiment with several million data points shows substantially better end-to-end QoS than is observed with pure IP routing. Pursuing the machine learning approach, we then address the challenging problem of dispatching incoming tasks to servers in Cloud systems so as to offer the best QoS and reliable job execution; an experimental system (the Task Allocation Platform) that we have developed is presented and used to compare several task allocation schemes, including a model driven algorithm, a reinforcement learning based scheme, and a ''sensible’’ allocation algorithm that assigns tasks to sub-systems that are observed to provide lower response time. These schemes are compared via measurements both among themselves and against a standard round-robin scheduler, with two architectures (with homogenous and heterogenous hosts having different processing capacities) and the conditions under which the different schemes offer better QoS are discussed. Since Cloud systems include both locally based servers at user premises and remote servers and multiple Clouds that can be reached over the Internet, we also describe a smart distributed system that combines local and remote Cloud facilities, allocating tasks dynamically to the service that offers the best overall QoS, and it includes a routing overlay which minimizes network delay for data transfer between Clouds. Internet-scale experiments that we report exhibit the effectiveness of our approach in adaptively distributing workload across multiple Clouds.

Web users can not be guaranteed that the results provided by Web search engines or recommender systems are either exhaustive or relevant to their search needs. Businesses have the commercial interest to rank higher on results or recommendations to attract more customers while Web search engines and recommender systems make their profit based on their advertisements. This research analyses the result rank relevance provided by the different Web search engines, metasearch engines, academic databases and recommender systems. We propose an Intelligent Search Assistant (ISA) that addresses these issues from the perspective of end-users acting as an interface between users and the different search engines; it emulates a Web Search Recommender System for general topic queries where the user explores the results provided. Our ISA sends the original query, retrieves the provided options from the Web and reorders the results. The proposed mathematical model of our ISA divides a user query into a multidimensional term vector. Our ISA is based on the Random Neural Network with Deep Learning Clusters. The potential value of each neuron or cluster is calculated by applying our innovative cost function to each snippet and weighting its dimension terms with different relevance parameters. Our ISA adapts to the perceived user interest learning user relevance on an iterative process where the user evaluates directly the listed results. Gradient Descent and Reinforcement Learning are used independently to update the Random Neural Network weights and we evaluate their performance based on the learning speed and result relevance. Finally, we present a new relevance metric which combines relevance and rank. We use this metric to validate and assess the learning performance of our proposed algorithm against other search engines. In some situations, our ISA and its iterative learning outperforms other search engines and recommender systems.

There are numerous security challenges in cellular mobile networks, many of which originate from the Internet world. One of these challenges is to answer the problem with increasing rate of signalling messages produced by smart devices. In particular, many services in the Internet are provided through mobile applications in an unobstructed manner, such that users get an always connected feeling. These services, which usually come from instant messaging, advertising and social networking areas, impose significant signalling loads on mobile networks by frequent exchange of control data in the background. Such services and applications could be built intentionally or unintentionally, and result in denial of service attacks known as signalling attacks or storms. Negative consequences, among others, include degradations of mobile network’s services, partial or complete net- work failures, increased battery consumption for infected mobile terminals. This thesis examines the influence of signalling storms on different mobile technologies, and proposes defensive mechanisms. More specifically, using stochastic modelling techniques, this thesis first presents a model of the vulnerability in a single 3G UMTS mobile terminal, and studies the influence of the system’s internal parameters on stability under a signalling storm. Further on, it presents a queueing network model of the radio access part of 3G UMTS and examines the effect of the radio resource control (RRC) inactivity timers. In presence of an attack, the proposed dynamic setting of the timers manage to lower the signalling load in the network and to increase the threshold above which a network failure could happen. Further on, the network model is upgraded into a more generic and detailed model, represent different generations of mobile technologies. It is than used to compare technologies with dedicated and shared organisation of resource allocation, referred to as traditional and contemporary networks, using performance metrics such as: signalling and communication delay, blocking probability, signalling load on the network’s nodes, bandwidth holding time, etc. Finally, based on the carried analysis, two mechanisms are proposed for detection of storms in real time, based on counting of same-type bandwidth allocations, and usage of allocated bandwidth. The mechanisms are evaluated using discrete event simulation in 3G UMTS, and experiments are done combining the detectors with a simple attack mitigation approach.

The increasing concentration of human populations in modern urbanised societies has aggravated the frequency and destruction of both natural and manmade disasters, and has motivated considerable research over the last few decades. Accompanying the development of computing technology, emergency navigation algorithms in built environment have evolved from off-line algorithms that direct evacuees in accordance with pre-deployed static evacuation plans to on-line algorithms that dynamically calculate egress paths for evacuees. However, these algorithms normally consider evacuees in a homogeneous manner, and ignore the different requirements and relative risk of death among different groups of people caused by different mobilities, physical strength, health conditions and level of resistance to hazard. Therefore, this work aims to develop systems and algorithms to dynamically customise distinct paths for different categories of evacuees. To this end, we borrow the concept of Cognitive Packet Network (CPN) and adapt it to the context of emergency navigation. On top of the CPN framework, we design several routing metrics to calculate distinct egress paths for different categories of evacuees. To improve the inter and intra-group coordination, several cooperative strategies are proposed to further optimise the routes calculated by the proposed routing algorithm. To provide a more accurate prediction to the congestion level of each egress path during an evacuation process under the effect of panic behaviours, we combine the CPN based routing algorithm with a G-network model to analyse the congestion level on a path via capturing the dynamics of diverse categories of evacuees under the influence of panic and re-routing decisions from the navigation system. Finally, we extend our work to large scale evacuations, and propose a G-network based emergency navigation algorithm to direct vehicles to safe areas in the aftermath of a large-scale disaster in an energy and time efficient manner.

Graph analytics is a fast growing and significant field in the visualization and data mining community, which is applied on numerous high-impact applications such as, network security, finance, and health care, providing users with adequate knowledge across various patterns within a given system. Although a series of methods have been developed in the past years for the analysis of unstructured collections of multi-dimensional points, graph analytics has only recently been explored. Despite the significant progress that has been achieved recently, there are still many open issues in the area, concerning not only the performance of the graph mining algorithms, but also producing effective graph visualizations in order to enhance human perception. The current thesis deals with the investigation of novel methods for graph analytics, in order to enhance data insight. Towards this direction, the current thesis proposes two methods so as to perform graph mining and visualization. Based on previous works related to graph mining, the current thesis suggests a set of novel graph features that are particularly efficient in identifying the behavioral patterns of the nodes on the graph. The specific features proposed, are able to capture the interaction of the neighborhoods with other nodes on the graph. Moreover, unlike previous approaches, the graph features introduced herein, include information from multiple node neighborhood sizes, thus capture long-range correlations between the nodes, and are able to depict the behavioral aspects of each node with high accuracy. Experimental evaluation on multiple datasets, shows that the use of the proposed graph features for the graph mining procedure, provides better results than the use of other state-of-the-art graph features. Thereafter, the focus is laid on the improvement of graph visualization methods towards enhanced human insight. In order to achieve this, the current thesis uses non-linear deformations so as to reduce visual clutter. Non-linear deformations have been previously used to magnify significant/cluttered regions in data or images for reducing clutter and enhancing the perception of patterns. Extending previous approaches, this work introduces a hierarchical approach for non-linear deformation that aims to reduce visual clutter by magnifying significant regions, and leading to enhanced visualizations of one/two/three-dimensional datasets. In this context, an energy function is utilized, which aims to determine the optimal deformation for every local region in the data, taking the information from multiple single-layer significance maps into consideration. The problem is subsequently transformed into an optimization problem for the minimization of the energy function under specific spatial constraints. Extended experimental evaluation provides evidence that the proposed hierarchical approach for the generation of the significance map surpasses current methods, and manages to effectively identify significant regions and deliver better results. The thesis is concluded with a discussion outlining the major achievements of the current work, as well as some possible drawbacks and other open issues of the proposed approaches that could be addressed in future works.