Distributed applications often make use of replicated state to afford a greater level of availability and throughput. This is achieved by allowing individual processes to progress without requiring prior synchronisation. This approach, termed optimistic replication, results in divergent replicas that must be reconciled to achieve an overall consistent state. Concurrent operations to shared objects in the replicas result in conflicting updates that require reconciliatory action to rectify. This typically takes the form of compensatory execution or simply undoing and rolling back client state. When considering user interaction with the application, there exists relationships and intent in the ordering and execution of these operations. The enactment of reconciliation that determines one action as conflicted may have far reaching implications with regards to the user’s original intent. In such scenarios, the compensatory action applied to a conflict may require previous operations to also be undone or compensated such that the user’s intent is maintained. Therefore, an ability to manage the contention to the shared data across the distributed application to pre-emptively lower conflicts resulting from these infringements is desirable. The aim is to not hinder throughput, achieved from the weaker consistency model known as eventual consistency. In this thesis, a model is presented for a contention management framework that schedules access using the expected execution inherent in the application domain to best inform the contention manager. A backoff scheme is employed to create an access schedule, preserving user intent for applications that require this high level of maintenance for user actions. By using such an approach, this results in a performance improvement seen in the reduction of the overall number of conflicts, while also improving overall system throughput. This thesis describes how the contention management scheme operates and, through experimentation, the performance benefits received.

From n-Tier client/server applications, to more complex academic Grids, or even the most recent and promising industrial Clouds, the last decade has witnessed significant developments in distributed computing. In spite of this conceptual heterogeneity, Service-Oriented Architecture (SOA) seems to have emerged as the common and underlying abstraction paradigm, even though different standards and technologies are applied across application domains. Suitable access to data and algorithms resident in SOAs via so-called ‘Science Gateways’ has thus become a pressing need in order to realize the benefits of distributed computing infrastructures. In an attempt to inform service-oriented systems design and developments in Grid-based biomedical research infrastructures, the applicant has consolidated work from three complementary experiences in European projects, which have developed and deployed large-scale production quality infrastructures and more recently Science Gateways to support research in breast cancer, pediatric diseases and neurodegenerative pathologies respectively. In analyzing the requirements from these biomedical applications the applicant was able to elaborate on commonly faced issues in Grid development and deployment, while proposing an adapted and extensible engineering framework. Grids implement a number of protocols, applications, standards and attempt to virtualize and harmonize accesses to them. Most Grid implementations therefore are instantiated as superposed software layers, often resulting in a low quality of services and quality of applications, thus making design and development increasingly complex, and rendering classical software engineering approaches unsuitable for Grid developments. The applicant proposes the application of a formal Model-Driven Engineering (MDE) approach to service-oriented developments, making it possible to define Grid-based architectures and Science Gateways that satisfy quality of service requirements, execution platform and distribution criteria at design time. An novel investigation is thus presented on the applicability of the resulting grid MDE (gMDE) to specific examples and conclusions are drawn on the benefits of this approach and its possible application to other areas, in particular that of Distributed Computing Infrastructures (DCI) interoperability, Science Gateways and Cloud architectures developments.

Grammar-oriented Object Design was shown to be a potent combination of extending methods, incorporating DSLs from a given business domain (BDSLs) and Variation-oriented Design in order to provide a seamless transition from business models to component-based software architectures. GOOD starts by extending current object modeling techniques to include the discovery and explicit modeling of higher levels of reuse, starting from subsystems, defining their manners using a domain-specific business language, i.e., using use-case gramars, that describe the rules governing the creation, dynamic configuration and collaboration of large-grained, business-process-scale, adaptive software components with pluggable behavior, through the application of architectural patterns and representation of component manners in the BDSL. 1his presents immense potential for applications in the domains of grid services, services on demand and a utility-based model of computing where a business need initiates the convergence of application components based on/from the manners of services they provide and require.

This thesis explores partially connected networks in associative memory tasks. The storage capacity of the fully connected associative net is reviewed. Simulation results demonstrate that the performance of this architecture was worse than that predicted by earlier analyses. New analysis is presented that accurately describes the behaviour of this net. A characterization of the storage capacity and information efficiency of the partially connected associative net is presented. In partially connected nets, one of the key problems is how to set the unit thresholds in such a way that the units which should not fire are quiet and those that should fire do so. New thresholding strategies are developed which enable this architecture to function well as an associative content-addressable memory. An important result is that the best thresholding strategies are functions of the firing history of an output unit and the number of active inputs impinging on it. Analysis is presented that predicts the behaviour of single and multi-layer partially connected nets using the best thresholding strategy. It shows that for a large range of parameters values the information efficiency of the partially connected net is greater than that of both the fully connected associative net and simple competitive nets.

With the variability in performance of the multitude of parallel environments available today, the conceptual overhead created by the need to anticipate runtime information to make design-time decisions has become overwhelming. Performance-critical applications and libraries carry implicit assumptions based on incidental metrics that are not portable to emerging computational platforms or even alternative contemporary architectures. Furthermore, the significance of runtime concerns such as makespan, energy efficiency and fault tolerance depends on the situational context. This thesis presents a case study in the application of both Mattson’s prescriptive pattern-oriented approach and the more principled structured parallelism formalism to the computational simulation of inelastic neutron scattering spectra on hybrid CPU/GPU platforms. The original ad hoc implementation as well as new patternbased and structured implementations are evaluated for relative performance and scalability. Two new structural abstractions are introduced to facilitate adaptation by lazy optimisation and runtime feedback. A deferred-choice abstraction represents a unified space of alternative structural program variants, allowing static adaptation through model-specific exhaustive calibration with regards to the extrafunctional concerns of runtime, average instantaneous power and total energy usage. Instrumented queues serve as mechanism for structural composition and provide a representation of extrafunctional state that allows realisation of a market-based decentralised coordination heuristic for competitive resource allocation and the Lyapunov drift algorithm for cooperative scheduling.

This thesis investigates techniques for the solution of partial differential equations (PDE) on advanced parallel architectures comprising central processing units (CPU) and graphics processing units (GPU). Many physical phenomena studied by scientists and engineers are modelled with PDEs, and these are often computationally expensive to solve. This is one of the main drivers of large-scale computing development. There are many well-established PDE solvers, however they are often inherently sequential. In consequence, there is a need to redesign the existing algorithms, and to develop new methods optimised for advanced parallel architectures. This task is challenging due to the need to identify and exploit opportunities for parallelism, and to deal with communication overheads. Moreover, a wide range of parallel platforms are available — interoperability issues arise if these are employed to work together. This thesis offers several contributions. First, performance characteristics of hybrid CPU-GPU platforms are analysed in detail in three case studies. Secondly, an optimised GPU implementation of the Preconditioned Conjugate Gradients (PCG) solver is presented. Thirdly, a multi-GPU iterative solver was developed — the Distributed Block Direct Solver (DBDS). Finally, and perhaps the most significant contribution, is the innovative streaming processing for FFT-based Poisson solvers. Each of these contributions offers significant insight into the application of advanced parallel systems in scientific computing. The techniques introduced in the case studies allow us to hide most of the communication overhead on hybrid CPU-GPU platforms. The proposed PCG implementation achieves 50–68% of the theoretical GPU peak performance, and it is more than 50% faster than the state-of-the-art solution (CUSP library). DBDS follows the Block Relaxation scheme to find the solution of linear systems on hybrid CPU-GPU platforms. The convergence of DBDS has been analysed and a procedure to compute a high-quality upper bound is derived. Thanks to the novel streaming processing technique, our FFT-based Poisson solvers are the first to handle problems larger than the GPU memory, and to enable multi- GPU processing with a linear speed-up. This is a significant improvement over the existing methods, which are designed to run on a single GPU, and are limited by the device memory size. Our algorithm needs only 6.9 seconds to solve a 2D Poisson problem with 2.4 billion variables (9 GB) on two Tesla C2050 GPUs (3 GB memory).

In 2006 NVIDIA introduced a new unified GPU architecture facilitating general-purpose computation on the GPU. The following year NVIDIA introduced CUDA, a parallel programming architecture for developing general purpose applications for direct execution on the new unified GPU. CUDA exposes the GPU's massively parallel architecture of the GPU so that parallel code can be written to execute much faster than its sequential counterpart. Although CUDA abstracts the underlying architecture, fully utilising and scheduling the GPU is non-trivial and has given rise to a new active area of research. Due to the inherent complexities pertaining to GPU development, in this thesis we explore and find efficient parallel mappings of existing and new parallel algorithms on the GPU using NVIDIA CUDA. We place particular emphasis on metaheuristics, image processing and designing reusable techniques and mappings that can be applied to other problems and domains. We begin by focusing on Ant Colony Optimisation (ACO), a nature inspired heuristic approach for solving optimisation problems. We present a versatile improved data-parallel approach for solving the Travelling Salesman Problem using ACO resulting in significant speedups. By extending our initial work, we show how existing mappings of ACO on the GPU are unable to compete against their sequential counterpart when common CPU optimisation strategies are employed and detail three distinct candidate set parallelisation strategies for execution on the GPU. By further extending our data-parallel approach we present the first implementation of an ACO-based edge detection algorithm on the GPU to reduce the execution time and improve the viability of ACO-based edge detection. We finish by presenting a new color edge detection technique using the volume of a pixel in the HSI color space along with a parallel GPU implementation that is able to withstand greater levels of noise than existing algorithms.

Data centres have been the primary focus of energy efficiency researches due to their expanding scales and increasing demands of energy. On the other hand, there are several orders of magnitude more end-users and personal computing devices worldwide. Even the modest energy savings from the users would scale up and yield significant impact. As a result, we take the approach towards energy-saving by working with the end-users. We recognise that users of ICT systems are often unaware of their power usage, and are therefore unable to take effective actions even if they wanted to save energy. Apart from energy awareness, the majority of end-users often lack of sufficient knowledge or skills to reduce their energy consumption while using computing devices. Moreover, there is no incentive for them to save energy, especially in public environments where they do not have financial responsibilities for their energy use. We propose a flexible energy monitor that gathers detailed energy usage across complex ICT systems, and provides end-users with accurate and timely feedback of their individual energy usage per workstation. We tailored our prototype energy monitor for a 2-year empirical study, with 83 student users of a university computer lab, and showed that end-users will change their use of computers to be more energy efficient, when sufficient feedback and incentives (rewards) are provided. In our measurements, weekly mean group power consumption as a whole reduced by up to 16%; and weekly individual user power usage reduced by up to 56% during active use. Based on our observations and collected data, we see possibilities of energy saving from both hardware and software components of personal computers. It requires coordination and collaboration between both system administrators and end-users to maximise energy savings. Institutional ‘green' policies are potentially helpful to enforce and regulate energy efficient use of ICT devices.

A brain-computer interface (BCI) measures brain activity and translates this activity into commands for a program to execute. BCls enable movement-free communication and interaction with technologies. This thesis evaluates the effectiveness and limitations of motion-onset visual evoked potentials (mVEP) based BCI as a control method for brain-computer games interaction. MVEP incorporates neural activity from the dorsal pathway of the visual system which allows more elegant visual stimuli than other types ofVEP and has yet to be used in computer games. This thesis investigates ifmVEP can be used as a control method in multiple computer games, what genre of game is best for interaction with m VEP and can we correct problems with existing VEP BCI computer games? Before conducting experiments involving games of different genres an evaluation of the present stateof- the-art BCI games was carried out in an extensive literature survey on BCI games categorised by genre. The literature survey shows that 'action' is the most popular genre in BCI games (49% of BC I games) and provides both games developers and BCI experts a set of design and development guidelines for BCI games. The conclusions of the survey led to the development of three BCI games of different genres namely action, puzzle and sports. The testing of different BCI games using a single paradigm enables thorough assessment ofmVEP as a control method. Five mVEP stimuli are presented as buttons to allow the subject to choose from five possible actions in each game. The performance was assessed based on offline and online BCI accuracy and game score. The results indicate that players could control the games with reasonable online accuracy (66% average for 5 class classification, with an average training accuracy of 74%). The next study intended on improving the initial study's results by adding the mVEP to an on screen HUD (Heads up Display), training in the same game environment as the participants are tested within and adding a questionnaire. Results indicate that the players could control the games with an average online accuracy of 71 %, a significant improvement from the previous study. After further analysis of recorded data the ideal setup for mVEP games is defined with key specifications indicating between three and four channels is most economical setup without influencing accuracy whilst averaging over three trials (minimises latency in communication). Finally, through the evaluation of a range o,fthe games related surveys, we found that players enjoyed the m VEP puzzle game most, rating it both the most enjoyable and appropriate game with m VEP control. Overall this thesis shows that m VEP can be used in multiple games genres with good accuracy and provides players with an entertaining and novel control method for computer games.

Cardiacvascular disease (CVD) is the single leading cause of death in the world, claiming 17.3 million lives a year according to the World Health Organisation (WHO). The development of magnetic resonance (MR) imaging has provided clinicians and researchers with effective tools to detect, assess and monitor the progress of the disease and treatments. MR imaging produces images with high spatial resolution using noninvasive and non-ionising techniques. However, quantitative analysis of the cardiovascular system from MR images remains challenging. The work presented in this thesis focuses on the utilization of cardiac motion information including motion tracking, quantification of the motion and prediction of clinical variables by incorporating motion information. The first main contributions of the thesis are approaches for sparse and dense motion tracking: a sparse set of key landmarks is detected and tracked. They are used as constraints to perform cardiac dense motion tracking using both 3D tagged and untagged image sequences from short-axis and long-axis MR views simultaneously. In order to improve speed and accuracy of the motion tracking, we also develop an approach to identify and track a sparse set of distinctive landmarks in the presence of relatively large deformations for myocardium motion tracking without applying dense motion tracking. An integrated framework is proposed to combine entropy and SVD-based sparse landmark detection with a MRF-based motion tracking framework. In addition, the regional wall thickness systolic dyssynchrony index (SDI) derived directly from sparse motion tracking provides accurate quantification of LV motion, which agrees well with the clinical measurements. In our last contribution, we successfully used manifold learning as a feature selection approach for a SVM-based classification and regression to analyse 209 cardiac MR image sequences. The SVM-based approaches directly operate on the manifold coordinates of the MR images without requiring any non-rigid registration or segmentation and is hence computationally efficient. We demonstrate that, by considering both inter- and intra-subject variation in the manifold learning, we are able to extract both anatomical and functional information. This can be used to construct powerful and reliable classifiers that are more predictive than global indices such as LV volume and mass. The manifold allows for investigating how much temporal information is needed improve the classification performance. The regression experiments demonstrate that there is a very strong correlation between manifold coordinates and obesity indices.