Thesis (M. Phil.)--Hong Kong University of Science and Technology, 2003. / Includes bibliographical references (leaves 69-70). Also available in electronic version. Access restricted to campus users.

Thesis (Ph. D.)--Ohio State University, 2003. / Title from first page of PDF file. Document formatted into pages; contains xvii, 172 p.; also includes graphics. Includes abstract and vita. Advisor: Dhabeleswar K. Panda, Dept. of Computer and Information Science. Includes bibliographical references (p. 167-172).

Thesis (M. Phil.)--University of Hong Kong, 2002. / Includes bibliographical references (leaves 65-66).

In this report, the feasibility of a massively parallel, nanoscale laser shock peening process is investigated. This report will give a fundamental background on laser shock peening processes in general. The background will include a description of the mechanisms associated with laser shock peening, and the theory behind laser shock peening. The experiments that were performed to develop a nanoscale laser shock peening process will also be described in detail. The problems associated with different experiments and the results will be presented. / text

Not available / text

Not available / text

Not available / text

Traditional cluster computing systems such as the supercomputers are equipped with specially designed high-performance hardware, which escalates the manufacturing cost and the energy cost of those systems. Due to such drawbacks and the diversified demand in computation, two new types of clusters are developed: the GPU clusters and the Hadoop clusters. The GPU cluster combines traditional CPU-only computing cluster with general purpose GPUs to accelerate the applications. Thanks to the massively-parallel architecture of the GPU, this type of system can deliver much higher performance-per-watt than the traditional computing clusters. The Hadoop cluster is another popular type of cluster computing system. It uses inexpensive off-the-shelf component and standard Ethernet to minimize manufacturing cost. The Hadoop systems are widely used throughout the industry. Alongside with the lowered cost, these new systems also bring their unique challenges. According to our study, the GPU clusters are prone to severe under-utilization due to the heterogeneous nature of its computation resources, and the Hadoop clusters are vulnerable to network congestion due to its limited network resources. In this research, we are trying to improve the throughput of these novel cluster computing systems by increasing the workload parallelism and network I/O parallelism.

Practical issues concerning the implementation and application of genetic algorithms to a number of optimisation problems are the main subjects dealt with in this thesis. Genetic algorithms (GAs) are an attractive class of computational models that attempt to mimic the mechanisms of natural evolution to solve problems in a wide variety of domains. A general purpose genetic algorithm toolkit is developed and applied to the Steiner Problem in Graphs and the Radio Link Frequency Assignment Problem. The toolkit is then extended to cover a large number of parallel genetic algorithm models which are then compared. Solutions for the two case studies are presented with each of the parallel GAs. The thesis begins with a general introduction to genetic algorithms. Holland's original genetic algorithm is described and it's workings illustrated on a simple function minimisation problem. The notion of a schema or similarity template as a basic building block in genetic algorithms is introduced and the schema theory presented. A description of important theoretical results is given and the introduction to genetic algorithms continues with practical issues that are dealt with in the second chapter. The basic components of a modern genetic algorithm are outlined and examples for important components, as found in the Jiterature, are given. The second chapter concludes with the description of a number of applications of genetic algorithms to areas such as function optimisation, combinatorial optimisation, genetic programming, process control and classifier systems. In Chapter 3, the sequential GA toolkit, GAmeter, is described. The General Search paradigm around which the toolkit is implemented is introduced. Notable characteristics of the genetic algorithms kernel and the user interface are mentioned. A popular function optimisation problem is used to illustrate important aspects of genetic algorithms and aspects specific to the toolkit. The Steiner Tree problem in graphs is the first of two case studies examined in detail in this thesis. This is a popular NP-complete problem with a range of applications in areas such as communications, scheduling and printed circuit design. A survey of standard techniques, such as simplification methods, exact algorithms and heuristics is given. Two possible representations for solving it using genetic algorithms are described and applied to a well-known set of problems. Chapter 4 concludes with a comparison of the best GA technique with other heuristics for this problem. The Radio Link Frequency Assignment Problem, described in Chapter 5, is the second case study investigated in this thesis. Genetic algorithms were applied to this problem as part of a EUCLID (European Cooperation for the Long Term in Defence) funded multi-national study to compare exact and heuristic techniques for hard combinatorial problems associated with military applications. A number of approaches used to solve this highly constrained, hard problem for genetic algorithms are described. These include a range of new genetic operators and catalytic terms that are added to the fitness function. Apart from the direct approach to solving this problem using genetic algorithms, for which the majority of operators and catalytic terms apply, an indirect approach which combines genetic algorithms with backtracking is described. The possibility of using a meta genetic algorithm to chose the best of a multitude of options, e.g. genetic operators and parameter settings for a GA applied to the Radio Link Frequency Assignment Problem is investigated. Results are reported for two sets of problems that were used by all participants in this project. An overview of the techniques investigated for this project is given and the chapter concludes with comparisons between all these techniques. In Chapter 6, an overview of general aspects in parallel processing is given. Parallel computer architectures, parallel programming paradigms and performance measurement are the main subjects dealt with in this chapter. Special emphasis is given to material relevant to the investigation on parallel genetic algorithms, presented in the following chapter. In Chapter 7, parallel genetic algorithms are examined in some detail. A number of parallel GA models are described and classified according to whether they are designed around the sequential GA or around a more natural model. A ParallelSequential General Search paradigm is presented that unifies the various parallel models and is used to extend the GA toolkit into a parallel GA toolkit for a parallel system based on Transputers. The parallel GA models are applied to problems from both of the case studies considered in this thesis. A comparison between the various parallel GA models concludes this chapter. The thesis finishes with a summary of a number of conclusions drawn from this research together with some suggestions for how this work may be continued in the future.

To make progress in the face of failures, long-running parallel applications need to save their state, known as a checkpoint. Unfortunately, current checkpointing techniques are becoming untenable on large-scale supercomputers. Many applications checkpoint all processes simultaneously--a technique that is easy to implement but often saturates the network and file system, causing a significant increase in checkpoint overhead. This thesis introduces compiler-assisted staggered checkpointing, where processes checkpoint at different places in the application text, thereby reducing contention for the network and file system. This checkpointing technique is algorithmically challenging since the number of possible solutions is enormous and the number of desirable solutions is small, but we have developed a compiler algorithm that both places staggered checkpoints in an application and ensures that the solution is desirable. This algorithm successfully places staggered checkpoints in parallel applications configured to use tens of thousands of processes. For our benchmarks, this algorithm successfully finds and places useful recovery lines that are up to 37% faster for all configurations than recovery lines where all processes write their data at approximately the same time. We also analyze the success of staggered checkpointing by investigating sets of application and system characteristics for which it reduces network and file system contention. We find that for many configurations, staggered checkpointing reduces both checkpointing time and overall execution time. To perform these analyses, we develop an event-driven simulator for large-scale systems that estimates the behavior of the network, global file system, and local hardware using predictive models. Our simulator allows us to accurately study applications that have thousands of processes; it on average predicts execution times as 83% of their measured value. / text