Curbing energy cravings in networks: A cross-sectional view across the micro-macro boundary

摘要

The soaring power dissipation of computing infrastructures has effectively become a key performance bottleneck. At the same time, thermal issues arising from power hungry devices result in compromising reliability of such systems. At present, the processor cores within a chip multiprocessor (CMP) are interconnected using on-chip networks. At the same time, the emerging era of cloud computing and the associated large-scale distributed computing model are stretching the limits of computer networking. Hence, as an essential and pervasive fabric that binds all computing machinery, there is an underlying network that appears with varying scale and characteristics. In this paper, we take a look at methodologies for better management of the power budget across the cross-section of computing systems traversing the boundary between micro-level on-chip networks to macro-level cloud computing. We discuss methodologies for power and thermal management both at micro and macro scales and explore the role of one in reducing the power budget of the other.