A framework for fully decentralised cycle stealing

Ordinary desktop computers continue to obtain ever more resources – in-creased processing power, memory, network speed and bandwidth – yet these resources spend much of their time underutilised. Cycle stealing frameworks harness these resources so they can be used for high-performance computing. Traditionally cycle stealing systems have used client-server based architectures which place significant limits on their ability to scale and the range of applica-tions they can support. By applying a fully decentralised network model to cycle stealing the limits of centralised models can be overcome. Using decentralised networks in this manner presents some difficulties which have not been encountered in their previous uses. Generally decentralised ap-plications do not require any significant fault tolerance guarantees. High-performance computing on the other hand requires very stringent guarantees to ensure correct results are obtained. Unfortunately mechanisms developed for traditional high-performance computing cannot be simply translated because of their reliance on a reliable storage mechanism. In the highly dynamic world of P2P computing this reliable storage is not available. As part of this research a fault tolerance system has been created which provides considerable reliability without the need for a persistent storage. As well as increased scalability, fully decentralised networks offer the ability for volunteers to communicate directly. This ability provides the possibility of supporting applications whose tasks require direct, message passing style communication. Previous cycle stealing systems have only supported embarrassingly parallel applications and applications with limited forms of communication so a new programming model has been developed which can support this style of communication within a cycle stealing context. In this thesis I present a fully decentralised cycle stealing framework. The framework addresses the problems of providing a reliable fault tolerance sys-tem and supporting direct communication between parallel tasks. The thesis includes a programming model for developing cycle stealing applications with direct inter-process communication and methods for optimising object locality on decentralised networks.

permGPU: Using graphics processing units in RNA microarray association studies.

BACKGROUND: Many analyses of microarray association studies involve permutation, bootstrap resampling and cross-validation, that are ideally formulated as embarrassingly parallel computing problems. Given that these analyses are computationally intensive, scalable approaches that can take advantage of multi-core processor systems need to be developed. RESULTS: We have developed a CUDA based implementation, permGPU, that employs graphics processing units in microarray association studies. We illustrate the performance and applicability of permGPU within the context of permutation resampling for a number of test statistics. An extensive simulation study demonstrates a dramatic increase in performance when using permGPU on an NVIDIA GTX 280 card compared to an optimized C/C++ solution running on a conventional Linux server. CONCLUSIONS: permGPU is available as an open-source stand-alone application and as an extension package for the R statistical environment. It provides a dramatic increase in performance for permutation resampling analysis in the context of microarray association studies. The current version offers six test statistics for carrying out permutation resampling analyses for binary, quantitative and censored time-to-event traits.

Execution of compute intensive applications on hybrid clouds (case study with mpiBLAST)

Cloud computing is the most recent realisation of computing as a utility. Recently, fields with substantial computational requirements, e.g., biology, are turning to clouds for cheap, on-demand provisioning of resources. Of interest to this paper is the execution of compute intensive applications on hybrid clouds. If application requirements exceed private cloud resource capacity, clients require scaling down their applications. The outcome of this research is Web technology realising a new form of cloud called HPC Hybrid Deakin (H2D) Cloud -- an experimental hybrid cloud capable of utilising both local and remote computational services for single large embarrassingly parallel applications.

Multi-cloud application execution and data management

Recently, fields with substantial computing requirementshave turned to cloud computing for economical, scalable, and on-demandprovisioning of required execution environments. However, current cloudofferings focus on providing individual servers while tasks such as applicationdistribution and data preparation are left to cloud users. This article presents anew form of cloud called HPC Hybrid Deakin (H2D) cloud; an experimentalhybrid cloud capable of utilising both local and remote computational servicesfor large embarrassingly parallel applications. As well as supporting execution,H2D also provides a new service, called DataVault, that provides transparentdata management services so all cloud-hosted clusters have required datasetsbefore commencing execution.

Distributed Feature Selection in Large n and Large p Regression Problems

Fitting statistical models is computationally challenging when the sample size or the dimension of the dataset is huge. An attractive approach for down-scaling the problem size is to first partition the dataset into subsets and then fit using distributed algorithms. The dataset can be partitioned either horizontally (in the sample space) or vertically (in the feature space), and the challenge arise in defining an algorithm with low communication, theoretical guarantees and excellent practical performance in general settings. For sample space partitioning, I propose a MEdian Selection Subset AGgregation Estimator ({\em message}) algorithm for solving these issues. The algorithm applies feature selection in parallel for each subset using regularized regression or Bayesian variable selection method, calculates the `median' feature inclusion index, estimates coefficients for the selected features in parallel for each subset, and then averages these estimates. The algorithm is simple, involves very minimal communication, scales efficiently in sample size, and has theoretical guarantees. I provide extensive experiments to show excellent performance in feature selection, estimation, prediction, and computation time relative to usual competitors.

While sample space partitioning is useful in handling datasets with large sample size, feature space partitioning is more effective when the data dimension is high. Existing methods for partitioning features, however, are either vulnerable to high correlations or inefficient in reducing the model dimension. In the thesis, I propose a new embarrassingly parallel framework named {\em DECO} for distributed variable selection and parameter estimation. In {\em DECO}, variables are first partitioned and allocated to m distributed workers. The decorrelated subset data within each worker are then fitted via any algorithm designed for high-dimensional problems. We show that by incorporating the decorrelation step, DECO can achieve consistent variable selection and parameter estimation on each subset with (almost) no assumptions. In addition, the convergence rate is nearly minimax optimal for both sparse and weakly sparse models and does NOT depend on the partition number m. Extensive numerical experiments are provided to illustrate the performance of the new framework.

For datasets with both large sample sizes and high dimensionality, I propose a new "divided-and-conquer" framework {\em DEME} (DECO-message) by leveraging both the {\em DECO} and the {\em message} algorithm. The new framework first partitions the dataset in the sample space into row cubes using {\em message} and then partition the feature space of the cubes using {\em DECO}. This procedure is equivalent to partitioning the original data matrix into multiple small blocks, each with a feasible size that can be stored and fitted in a computer in parallel. The results are then synthezied via the {\em DECO} and {\em message} algorithm in a reverse order to produce the final output. The whole framework is extremely scalable.

Cognitive assessment in closed head injury : stability, validity and parallel forms for two neuropsychological measures of recovery

Investigated the psychometric properties of the original and alternate sets of the Trail Making Test (TMT) and the Controlled Oral Word Association Test (COWAT; A. L. Benton and D. Hamsher, 1978) in 50 orthopedic and 15 closed head injured (1 yr after trauma) patients (aged 15–59 yrs). Although the alternate forms of both measures proved to be stable and consistent with each other in both groups, only the parallel sets of TMT reliably discriminated the clinical group from controls. Practice effects in the head injured were significant only for Trail B of TMT. Factor analysis of the control group's results identified Verbal Knowledge as a major contributor to performance on COWAT, whereas TMT was more dependent on Rapid Visual Search and Visuomotor Sequencing.

Scheduling trains as a blocking parallel-machine job shop scheduling problem

In this paper, the train scheduling problem is modelled as a blocking parallel-machine job shop scheduling (BPMJSS) problem. In the model, trains, single-track sections and multiple-track sections, respectively, are synonymous with jobs, single machines and parallel machines, and an operation is regarded as the movement/traversal of a train across a section. Due to the lack of buffer space, the real-life case should consider blocking or hold-while-wait constraints, which means that a track section cannot release and must hold the train until next section on the routing becomes available. Based on literature review and our analysis, it is very hard to find a feasible complete schedule directly for BPMJSS problems. Firstly, a parallel-machine job-shop-scheduling (PMJSS) problem is solved by an improved shifting bottleneck procedure (SBP) algorithm without considering blocking conditions. Inspired by the proposed SBP algorithm, feasibility satisfaction procedure (FSP) algorithm is developed to solve and analyse the BPMJSS problem, by an alternative graph model that is an extension of the classical disjunctive graph models. The proposed algorithms have been implemented and validated using real-world data from Queensland Rail. Sensitivity analysis has been applied by considering train length, upgrading track sections, increasing train speed and changing bottleneck sections. The outcomes show that the proposed methodology would be a very useful tool for the real-life train scheduling problems

Theoretical investigation of thermal tweezers for parallel manipulation of atoms and nanoparticles on surfaces

A major focus of research in nanotechnology is the development of novel, high throughput techniques for fabrication of arbitrarily shaped surface nanostructures of sub 100 nm to atomic scale. A related pursuit is the development of simple and efficient means for parallel manipulation and redistribution of adsorbed atoms, molecules and nanoparticles on surfaces – adparticle manipulation. These techniques will be used for the manufacture of nanoscale surface supported functional devices in nanotechnologies such as quantum computing, molecular electronics and lab-on-achip, as well as for modifying surfaces to obtain novel optical, electronic, chemical, or mechanical properties. A favourable approach to formation of surface nanostructures is self-assembly. In self-assembly, nanostructures are grown by aggregation of individual adparticles that diffuse by thermally activated processes on the surface. The passive nature of this process means it is generally not suited to formation of arbitrarily shaped structures. The self-assembly of nanostructures at arbitrary positions has been demonstrated, though these have typically required a pre-patterning treatment of the surface using sophisticated techniques such as electron beam lithography. On the other hand, a parallel adparticle manipulation technique would be suited for directing the selfassembly process to occur at arbitrary positions, without the need for pre-patterning the surface. There is at present a lack of techniques for parallel manipulation and redistribution of adparticles to arbitrary positions on the surface. This is an issue that needs to be addressed since these techniques can play an important role in nanotechnology. In this thesis, we propose such a technique – thermal tweezers. In thermal tweezers, adparticles are redistributed by localised heating of the surface. This locally enhances surface diffusion of adparticles so that they rapidly diffuse away from the heated regions. Using this technique, the redistribution of adparticles to form a desired pattern is achieved by heating the surface at specific regions. In this project, we have focussed on the holographic implementation of this approach, where the surface is heated by holographic patterns of interfering pulsed laser beams. This implementation is suitable for the formation of arbitrarily shaped structures; the only condition is that the shape can be produced by holographic means. In the simplest case, the laser pulses are linearly polarised and intersect to form an interference pattern that is a modulation of intensity along a single direction. Strong optical absorption at the intensity maxima of the interference pattern results in approximately a sinusoidal variation of the surface temperature along one direction. The main aim of this research project is to investigate the feasibility of the holographic implementation of thermal tweezers as an adparticle manipulation technique. Firstly, we investigate theoretically the surface diffusion of adparticles in the presence of sinusoidal modulation of the surface temperature. Very strong redistribution of adparticles is predicted when there is strong interaction between the adparticle and the surface, and the amplitude of the temperature modulation is ~100 K. We have proposed a thin metallic film deposited on a glass substrate heated by interfering laser beams (optical wavelengths) as a means of generating very large amplitude of surface temperature modulation. Indeed, we predict theoretically by numerical solution of the thermal conduction equation that amplitude of the temperature modulation on the metallic film can be much greater than 100 K when heated by nanosecond pulses with an energy ~1 mJ. The formation of surface nanostructures of less than 100 nm in width is predicted at optical wavelengths in this implementation of thermal tweezers. Furthermore, we propose a simple extension to this technique where spatial phase shift of the temperature modulation effectively doubles or triples the resolution. At the same time, increased resolution is predicted by reducing the wavelength of the laser pulses. In addition, we present two distinctly different, computationally efficient numerical approaches for theoretical investigation of surface diffusion of interacting adparticles – the Monte Carlo Interaction Method (MCIM) and the random potential well method (RPWM). Using each of these approaches we have investigated thermal tweezers for redistribution of both strongly and weakly interacting adparticles. We have predicted that strong interactions between adparticles can increase the effectiveness of thermal tweezers, by demonstrating practically complete adparticle redistribution into the low temperature regions of the surface. This is promising from the point of view of thermal tweezers applied to directed self-assembly of nanostructures. Finally, we present a new and more efficient numerical approach to theoretical investigation of thermal tweezers of non-interacting adparticles. In this approach, the local diffusion coefficient is determined from solution of the Fokker-Planck equation. The diffusion equation is then solved numerically using the finite volume method (FVM) to directly obtain the probability density of adparticle position. We compare predictions of this approach to those of the Ermak algorithm solution of the Langevin equation, and relatively good agreement is shown at intermediate and high friction. In the low friction regime, we predict and investigate the phenomenon of ‘optimal’ friction and describe its occurrence due to very long jumps of adparticles as they diffuse from the hot regions of the surface. Future research directions, both theoretical and experimental are also discussed.