Green HPC: Optimizing Software Stack Energy Efficiency of Large Data Systems

High-performance computing (HPC) is indispensable in modern scientific research and industry applications, but its energy consumption is a growing concern. This thesis presents two novel approaches to optimize energy consumption in large data systems. The first chapter of the thesis will discuss the use of Dynamic Voltage and Frequency Scaling (DVFS) to optimize the energy efficiency of two popular lossy compression algorithms: SZ and ZFP. By adjusting the voltage and frequency levels of computing resources, DVFS can reduce energy consumption while maintaining the desired level of performance and accuracy. The second chapter of the thesis will focus on a detailed comparison and analysis of asynchronous and synchronous checkpointing energy consumption using the VELOC and GenericIO libraries. The study investigates the trade-offs between these two checkpointing techniques, offering insights into their energy consumption patterns and performance impacts on large-scale HPC systems. Based on the analysis, we provide recommendations for choosing the most energy-efficient checkpointing method for specific application scenarios. Together, these two approaches contribute to the development of Green HPC, paving the way for more sustainable and energy-efficient large data systems. This thesis will provide valuable insights for researchers and industry practitioners aiming to optimize energy consumption while maintaining high-performance computing capabilities. i