OPTIMIZING PROCESSOR AND MEMORY FOR GREEN COMPUTING

Abstract

Energy efficiency has become one of the most important factors in the development of computer systems. Increasingly power-hungry processors and memory subsystem have reinforced the need for aggressive power management. Dynamic voltage scaling has become a common consideration for designing energy efficient CPUs in systems ranging from portable devices to large-scale systems. As applications become more data centric and put more pressure on memory subsystem, managing energy consumption of main memory is also becoming critical. Subsequently in this dissertation, we address the issues in designing energy efficient CPU and memory for personal computing devices as well as large-scale systems.For large-scale systems, we address memory subsystem dedicated to buffer cache which accounts for the majority of memory usage in server environment. We take advantage of the I/O handling routines in the OS kernel to hide the delay incurred by the memory state transition so that performance degradation is minimized while high energy savings is achieved. We also address interactive workloads, which account for the bulk of the processing demand on modern mobile or desktop systems. We propose Interaction-Aware Dynamic Voltage Scaling (IADVS) for CPU and Interaction-Aware Memory Energy Management (IAMEM) for memory. The IA framework relies on automatic correlation of user-initiated tasks with the demand placed on CPU and memory to accurately predict power states for CPU and memory. Both mechanisms achieve maximal energy savings while minimizing the impact on the application's performance.