A First-Principles Approach to Performance, Power, and Energy Models for Contemporary Multi- and Many-Core Processors

En savoir plus sur le livre

Models are essential in engineering, serving as critical tools in scientific and high-performance computing. Performance models evaluate implementation efficiency, identify bottlenecks, and guide optimization strategies. Similarly, power and energy models address demands across computing, from extending battery life in mobile devices to managing sustainable power budgets for advanced supercomputers. This thesis encompasses two primary efforts. The first involves developing performance, power, and energy models specifically for loop-based streaming codes. These models are grounded in first principles and specifications, enhanced by empirical data to achieve unprecedented accuracy. The design emphasizes generality by considering various processors from vendors like Intel, AMD, and IBM. The second effort aims to deepen the understanding of the performance, power, and energy behaviors of modern server processors through insights derived from the models' foundational principles. Key scientific contributions include identifying universal behaviors, uncovering governing mechanisms, and establishing best practices for optimizing performance alongside power and energy consumption.