Power-Aware Multicore Processing

点击翻译按钮获取中文摘要

The goal of this proposal is to develop techniques to reduce power dissipation in multicore processors. Multicore processors, also referred to as chip multiprocessors (or simply CMPs), are employed in a wide range of computing systems from data centers to desktops and embedded processors. The energy consumed by such computing systems accounts for a considerable share of our daily energy consumption. Therefore, developing power efficient multicore processors is a critical aspect of green computing. CMPs achieve high performance by using multiple processor cores on a single chip. While this approach has proven to provide effective means for current technologies, there are several studies that suggest that future CMPs will become power limited. Consequently, providing enough power for the chip will require turning off some sections. One way to reduce the proportion of these unused sections (also referred to as dark silicon) is to develop low-power CMPs. We will investigate solutions for enhancing power efficiency in future CMPs. To achieve this goal we suggest developing alternative power-aware designs. We will investigate a number of issues influencing the overall power dissipation in CMPs. We will study power scalability and the parameters contributing to the excessive power dissipation growth in large-scale CMPs. We will also analyze resource utilization in different CMP designs and identify resource utilization inefficiencies. We will use our findings to redesign conventional CMPs for better power efficiency. Our proposed research makes important contributions to Canadian society. First, the proposed research program aims at developing "greener" computing infrastructures where power hungry processors are replaced with low-power alternatives. Second, the resulting knowledge belongs to an area critical to Canada's future leadership in advanced technologies. Many Canadian companies can benefit from the findings of the proposed research and the resulting HQP training.