No description available.

Power, performance, and cost dictate the procurement and operation of high-performance computing (HPC) systems. These systems use graphics processing units (GPUs) for performance boost. In order to identify inexpensive-to-acquire and inexpensive-to-operate systems, it is important to do a systematic comparison of such systems with respect to power, performance and energy characteristics with the end use applications. Additionally, the chosen systems must often achieve performance objectives without exceeding their respective power budgets, a task that is usually borne by a software-based power management system. Accurately predicting the power consumption of an application at different DVFS levels (or more generally, different processor configurations) is paramount for the efficient functioning of such a management system. This thesis intends to apply the latest in the state-of-the-art in green computing research to optimize the total cost of acquisition and ownership of heterogeneous computing systems. To achieve this we take a two-fold approach. First, we explore the issue of greener device selection by characterizing device power and performance. For this, we explore previously untapped opportunities arising from a special type of graphics processor --- the low-power integrated GPU --- which is commonly available in commodity systems. We compare the greenness (power, energy, and energy-delay product $rightarrow$ EDP) of the integrated GPU against a CPU running at different frequencies for the specific application domain of scientific visualization. Second, we explore the problem of predicting the power consumption of a GPU at different DVFS states via machine-learning techniques. Specifically, we perform statistically rigorous experiments to uncover the strengths and weaknesses of eight different machine-learning techniques (namely, ZeroR, simple linear regression, KNN, bagging, random forest, SMO regression, decision tree, and neural networks) in predicting GPU power consumption at different frequencies. Our study shows that a support vector machine-aided regression model (i.e., SMO regression) achieves the highest accuracy with a mean absolute error (MAE) of 4.5%. We also observe that the random forest method produces the most consistent results with a reasonable overall MAE of 7.4%. Our results also show that different models operate best in distinct regions of the application space. We, therefore, develop a novel, ensemble technique drawing the best characteristics of the various algorithms, which reduces the MAE to 3.5% and maximum error to 11% from 20% for SMO regression. / MS

No description available.

No description available.

Today, power consumption costs supercomputer centers millions of dollars annually and the heat produced can reduce system reliability and availability. Achieving high performance while reducing power consumption is challenging since power and performance are inextricably interwoven; reducing power often results in degradation in performance. This thesis aims to address these challenges by providing theories, techniques, and tools to 1) accurately predict performance and improve it in systems with advanced hierarchical memories, 2) understand and evaluate power and its impacts on performance, 3) control power and performance for maximum efficiency. Our theories, techniques, and tools have been applied to high-end computing systems. Our theroetical models can improve algorithm performance by up to 59% and accurately predict the impacts of power on performance. Our techniques can evaluate power consumption of high-end computing systems and their applications with fine granularity and save up to 36% energy with little performance degradation. / Ph. D.

The object of this research is to optimize the overall power system performance using FACTS devices. Particularly, it is intended to improve the reliability, and the performance of the power system considering steady state operating condition as well as the system subjected to small and large disturbances. The methodology proposed to achieve this goal corresponds to an enhanced particle swarm optimizer (Enhanced-PSO) that is proven in this work to have several advantages, in terms of accuracy and computational effort, as compared with other existing methods. Once the performance of the Enhanced PSO is verified, a multi-stage PSO-based optimization framework is proposed for optimizing the power system reliability (N-1 contingency criterion). The algorithm finds optimal settings for present infrastructure (generator outputs, transformers tap ratios and capacitor banks settings) as well as optimal control references for distributed static series compensators (DSSC) and optimal locations, sizes and control settings for static compensator (STATCOM) units. Finally, a two-stage optimization algorithm is proposed to improve the power system performance in steady state conditions and when small and large perturbations are applied to the system. In this case, the algorithm provides optimal control references for DSSC modules, optimal location and sizes for capacitor banks, and optimal location, sizes and control parameters for STATCOM units (internal and external controllers), so that the loadability and the damping of the system are maximized at minimum cost. Simulation results throughout this research show a significant improvement of the power system reliability and performance after the system is optimized.

Wave energy conversion technology has gained popularity due to its potential to be-come one of the most preferred energy sources. Its high energy density and low car-bon footprint have inspired the development of many wave energy converter (WEC) technologies, few of which have made their way to commercialisation, and many are progressing. The Floating Power Plant (FPP) device is a combined floating wind and wave converter. The company, Floating Power Plant, was established in 2004 and has developed and patented a floating device that consists of a semi-submersible that serves as a foundation for a single wind turbine and hosts four wave energy converters (WECs). Each WEC consists of a partially submerged wave absorber whose pitching motion generates energy from incoming waves. The wave absorbers are connected to an oil hydraulic power take-off system located in a dry “engine room” above the free water surface, where the mechanical energy in the absorber is converted to electricity. When undergoing pitching movements, there are interactions between individual wave absorbers and the surrounding platform. This thesis focuses on developing methods to improve the FPP WEC’s hydrodynamic interactions. The first part of this thesis optimises the wave absorber (WA) ballast. An ana-lytical model is developed to enable systematic selection of WA ballast combination with significantly less computational effort when compared with the more conven-tional means, such as using CAD software. The study suggests an algorithm with which the absorbed power and resonance frequency can be improved and adjusted by manipulating the ballasts’ mass, the position of its centre of gravity, placement and inclination of the WA. The proposed method is generic and can be applied to other WEC concepts or submerged bodies in general. The results show the feasibility of designing the absorber ballast to offer passive control for increased wave absorption. It demonstrates the effect of ballast on the WA inclination, resonance frequency and response amplitude operator (RAO). The second part focuses on the optimisation of the FPP platform geometry. The genetic algorithm optimisation technique is implemented to maximise the annual en-ergy produced by the relative pitch motion of the WA to the floating platform. The optimised variables are characteristic lengths of the floating platform, most of which are part of the immediate surrounding walls of the absorber. The objective function is a function of the WA’s annual energy production (AEP) and RAO. Results show the feasibility of improving the hydrodynamic interaction between the floating platform and its integrated wave absorbers for a given wave climate by using a heuristic search technique. The number of iterations to convergence tends towards increased values when considering more optimised variables. It is also observed that the computational time appears to be independent of the number of variables but is significantly impacted by the computational power of the machine used.