Search

In large-scale computer systems and software development, model-driven engineering is an approach that focuses on the development and management of models. The models are usually expressed in diagrams, textual notations, or code. Most of these models persist in state-based formats. While state-based persistence has certain advantages, it is problematic when it comes to detecting changes in large-scale models. As an alternative, this work proposes a change-based approach that involves persisting the full sequence of changes made to models. Persisting a model in a change-based format has the potential to deliver benefits over state-based persistence, such as the ability to perform model differencing and conflict detection much faster and more precisely. This can then yield positive follow-on effects to help developers compare and merge models in collaborative modelling environments. Nevertheless, change-based persistence also comes with downsides, including increased model loading time. This work investigates two approaches to reduce loading time. The first is to identify and ignore superseded changes, and the second uses hybrid model persistence. While the former is still greatly outperformed by loading models from state-based persistence, the latter experiences only a slight slowdown in most cases. This work also proposes an approach for faster model differencing and conflict detection. It works by exploiting the nature of change-based persistence, which allows finding differences and conflicts between two versions of a model by comparing only the last set of changes applied to them, without having to compare every element and feature in both versions as is traditionally done in state-based model comparison. This work's evaluation shows that the proposed change-based model differencing and conflict detection outperform the existing traditional state-based approach. Nevertheless, models that have been excessively modified could impair the performance of the proposed model differencing and conflict detection as numerous change records must be loaded into memory.

Solving real-world combinatorial problems is involved in many industry fields to minimise operational cost or to maximise profit, or both. Along with continuous growth in computing power, many asset management decision-making processes that were originally solved by hand now tend to be based on big data analysis. Larger scale problem can be solved and more detailed operation instructions can be delivered. In this thesis, we investigate models and algorithms to solve large scale Geographically Distributed asset Maintenance Problems (GDMP). Our study of the problem was motivated by our business partner, Gaist solutions Ltd., to optimise scheduling of maintenance actions for a drainage system in an urban area. The models and solution methods proposed in the thesis can be applied to many similar issues arising in other industry fields. The thesis contains three parts. We firstly built a risk driven model considering vehicle routing problems and the asset degradation information. A hyperheuristic method embedded with customised low-level heuristics is employed to solve our real-world drainage maintenance problem in Blackpool. Computational results show that our hyperheuristic approach can, within reasonable CPU time, produce much higher quality solutions than the scheduling strategy currently implemented by Blackpool council. We then attempt to develop more efficient solution approaches to tackle our GDMP. We study various hyperheuristics and propose efficient local search strategies in part II. We present computational results on standard periodic vehicle routing problem instances and our GDMP instances. Based on manifold experimental evidences, we summarise the principles of designing heuristic based solution approaches to solve combinatorial problems. Last bu not least, we investigate a related decision making problem from highway maintenance, that is again of interest to Gaist solutions Ltd. We aim to make a strategical decision to choose a cost effective method of delivering the road inspection at a national scale. We build the analysis based on the Chinese Postman Problem and theoretically proof the modelling feasibility in real-world road inspection situations. We also propose a novel graph reduction process to allow effective computation over very large data sets.

This thesis presents my work on the pickup and multiple delivery problem, a real-world vehicle routing and scheduling problem with soft time windows, working time and last-in-first-out constraints, developed in collaboration with Transfaction Ltd., who conduct logistics analysis for several large retailers in the UK. A summary of relevant background literature is presented highlighting where my research fits into and contributes to the broader academic landscape. I present a detailed model of the problem and thoroughly analyse a case-study data set, obtaining distributions used for further research. A new variable neighbourhood descent with memory hyper-heuristic is presented and shown to be an effective technique for solving instances of the real-world problem. I analyse strategies for cooperation and competition amongst haulage companies and quantify their effectiveness. The value of time and timely information for planning pickup and delivery requests is investigated. The insights gained are of real industrial relevance, highlighting how a variety of business decisions can produce significant cost savings.

The technological dependency of products or services provided by a particular cloud platform or provider (i.e. cloud vendor lock-in) leaves cloud users unprotected against service failures and providers going out of business, and unable to modernise their software applications by exploiting new technologies and cheaper services from alternative clouds. High portability is key to ensure a smooth migration of software applications between clouds, reducing the risk of vendor lock-in. This research identifies and models key factors that impact the portability of legacy web applications in cloud computing. Unlike existing cloud portability studies, we use a combination of techniques from empirical software engineering, software quality and areas related to cloud, including service-oriented computing and distributed systems, to carry out a rigorous experimental study of four factors impacting on cloud application portability. In addition, we exploit established methods for software effort prediction to build regression models for predicting the effort required to increase cloud application portability. Our results show that software coupling, authentication technology, cloud platform and service are statistically significant and scientifically relevant factors for cloud application portability in the experiments undertaken. Furthermore, the experimental data enabled the development of fair (mean magnitude of relative error, MMRE, between 0.493 and 0.875), good (MMRE between 0.386 and 0.493) and excellent (MMRE not exceeding 0.368) regression models for predicting the effort of increasing the portability of legacy cloud applications. By providing empirical evidence of factors that impact cloud application portability and building effort prediction models, our research contributes to improving decision making when migrating legacy applications between clouds, and to mitigating the risks associated with cloud vendor lock-in.

The human immune system has characteristics such as self-organisation, robustness and adaptivity that may be useful in the development of adaptive systems. One suitable application area for adaptive systems is Information Filtering (IF). Within the context of IF, learning and adapting user profiles is an important research area. In an individual profile, an IF system has to rely on the ability of the user profile to maintain a satisfactory level of filtering accuracy for as long as it is being used. This thesis explores a possible way to enable Artificial Immune Systems (AIS) to filter information in the context of profile adaptation. Previous work has investigated this issue from the perspective of self-organisation based on Autopoetic Theory. In contrast, this current work approaches the problem from the perspective of diversity inspired by the concept of dynamic clonal selection and gene library to maintain sufficient diversity. An immune inspired IF for profile adaptation is proposed and developed. This algorithm is demonstrated to work in detecting relevant documents by using a single profile to recognize a user’s interests and to adapt to changes in them. We employed a virtual user tested on a web document corpus to test the profile on learning of an emerging new topic of interest and forgetting uninteresting topics. The results clearly indicate the profile’s ability to adapt to frequent variations and radical changes in user interest. This work has focused on textual information, but it may have the potential to be applied in other media such as audio and images in which adaptivity to dynamic environments is crucial. These are all interesting future directions in which this work might develop.

In Model-Driven Engineering, model transformation is a key model management operation, used to translate models between notations. Model transformation can be used for many engineering activities, for instance as a preliminary to merging models from different meta- models, or to generate codes from diagrammatic models. A mapping model needs to be developed (the transformation specification) to represent relations between concepts from the metamodels. The evaluation of the mapping model creates new challenges, for both conventional verification and validation, and also in guaranteeing that models generated by applying the transformation specification to source models still retain the intention of the initial transformation requirements. Most model transformation creates and evaluates a transformation specification in an ad-hoc manner. The specifications are usu- ally unstructured, and the quality of the transformations can only be assessed when the transformations are used. Analysis is not systematically applied even when the transformations are in use, so there is no way to determine whether the transformations are correct and consistent. This thesis addresses the problem of systematic creation and analysis of model transformation, via a facility for planning and designing model transformations which have conceptual-level properties that are tractable to formal analysis. We proposed a framework that provides steps to systematically build a model transformation specification, a visual notation for specifying model transformation and a template-based approach for producing a formal specification that is not just structure-equivalent but also amenable to formal analysis. The framework allows evaluation of syntactic and semantic correctness of generated models, metamodel coverage, and semantic correctness of the transformations themselves, with the help of snapshot analysis using patterns.

Model-Driven Engineering (MDE) and Search-Based Software Engineering (SBSE) are development approaches that focus on automation to increase productivity and throughput. MDE focuses on high-level domain models and the automatic management of models to perform development processes, such as model validation or code generation. SBSE on the other hand, treats software engineering problems as optimisation problems, and aims to automatically discover solutions rather than systematically construct them. SBSE techniques have been shown to be beneficial at all stages in the software development life-cycle. There has, however, been few attempts at applying SBSE techniques to the MDE domain and all are problem-specific. In this thesis we present a method of encoding MDE models that enables many robust SBSE techniques to be applied to a wide-range of MDE problems. We use the model representation to address three in-scope MDE problems: discovering an optimal domain model; extracting a model of runtime system behaviour; and applying sensitivity analysis to model management operations in order to analyse the uncertainty present in models. We perform an empirical analysis of two important properties of the representation, locality and redundancy, which have both been shown to affect the ability of SBSE techniques to discover solutions, and we propose a detailed plan for further analysis of the representation, or other representations of its kind.

Scientific research is a vanguard domain of human activity. Researchers aim for a systematic, objective approach, but also for pushing forward the boundaries of knowledge through the use of ever-more advanced instruments and techniques. Computer simulations represent such scientific instruments, capable of harvesting information in response to questions beyond the scope of traditional experimental techniques. The benefits of using them must, however, be considered together with aspects that led to criticism and lack of confidence e.g. they are difficult to analyze and validate, assumptions are only partially managed. This thesis scopes down the vast domain of simulation-based research, to the use of agent-based simulations for studying complex systems. The use of structured argumentation in a scientific research context is studied as a means for addressing the core limitations of simulation-based research. The Goal Structuring Notation has been used effectively in its originating domain - Safety Critical Systems - in addressing similar problems to simulation-based research. Through the use of this notation, this research emphasizes the difficulty of expressing compelling arguments, even in journal publications; in addition, it propose a set of extensions to the notation in order to adapt it to the scientific discourse. Finally, it shows the implications of studying a model in a rigorous, exhaustive manner, over the claims that can be made through it.

Enterprises are organisations with multiple business processes; they often use Enterprise Information Systems (EIS) to support these business processes. The concept of an EIS has arisen from the need to deal with the increasingly volatile requirements of modern large{scale organisations. EIS are growing in use and are now being used to support government, health care, and non-profit / non-governmental organisations. The development of EIS has been affected significantly by the complexity and size of enterprises and their business processes, in addition to the influences of economical, social, and governmental factors. There are many challenges associated with building EIS. Three critical ones identied in the literature are: adequately satisfying organisational requirements; building valid and stakeholder-acceptable business processes; and providing repeatable and rigorous approaches to establish shared understanding of EIS goals. These challenges are difficult to cope with because of the need to deal with different goals, changes in goals, and the problem of how to transform these goals into system requirements and, ultimately, to an EIS architecture. This thesis contributes a rigorous approach for identifying and describing the enterprise-level requirements of IT developers, managers, and other stakeholders of an enterprise. The approach provides two modelling and tool-supported processes to help establish a rigorous model of EIS goals. It also provides support for transforming goals to a strategic EIS architecture. The approach presented in the thesis is based on the concepts of Goal-oriented software engineering. The thesis presents a novel Process Model, KAOS-B that extends goal-oriented software engineering approaches with new concepts and techniques for EIS. Further, to support the transition from requirements to an EIS architecture, an EIS Architecture Process Model (EAPM), is designed and evaluated. Using KAOS-B and EAPM in concert provides a rigorous, repeatable and tool-supported approach for analysing, and designing a strategic EIS architecture. The thesis illustrates the approach with two substantial examples from the health informatics and critical systems domain.

Software changes over time. During the lifetime of a software system, unintended behaviour must be corrected and new requirements satisfied. Because software changes are costly, tools for automatically managing change are commonplace. Contemporary software development environments can automatically perform change management tasks such as impact analysis, refactoring and background compilation. Increasingly, models and modelling languages are first-class citizens in software development. Model-Driven Engineering (MDE), a state-of-the-art approach to software engineering, prescribes the use of models throughout the software engineering process. In MDE, modelling tools and task-specific language are used to generate an ultimate artefact, such as simulation models or working code. Contemporary MDE environments provide little support for managing a type of evolution termed model-metamodel co-evolution, in which changes to a modelling language are propagated to models. This thesis demonstrates that model-metamodel co-evolution occurs often in MDE projects, and that dedicated structures and processes for its management can increase developer productivity. Structures and processes for managing model-metamodel co-evolution are proposed, developed, and then evaluated by comparison to existing structures and processes with quantitative and qualitative techniques.

Emergent systems are often thought of as special, and are often linked to desirable properties like robustness, fault tolerance and adaptability. But, though not well understood, emergence is not a magical, unfathomable property. We introduce neutral emergence as a new way to explore emergent phenomena, showing that being good enough, enough of the time may actually yield more robust solutions more quickly. We then use cellular automata as a substrate to investigate emergence, and find they are capable of exhibiting emergent phenomena through coarse graining. Coarse graining shows us that emergence is a relative concept - while some models may be more useful, there is no correct emergent model - and that emergence is lossy, mapping the high level model to a subset of the low level behaviour. We develop a method of quantifying the 'goodness' of a coarse graining (and the quality of the emergent model) and use this to find emergent models - and, later, the emergent models we want - automatically.