CAREER: Self-Regulating Power-Aware Interconnection Networks

Princeton University, Princeton NJ

Investigators

Abstract

Power is becoming increasingly as important, if not more important than performance in many digital systems ranging from PCs and servers to Internet routers and embedded systems-on-a-chip. While there has been substantial research exploring the power efficiency of the processing and memory elements of digital systems, research investigating the power consumption of communication elements has been lagging. As a wide range of digital systems becomes increasingly interconnected, it is now both timely and critical to explore power-aware interconnection networks. In this proposal, we outline our plans to research, develop and build self-regulating power-aware interconnection networks that trade off power and performance automatically while meeting design constraints. In these networks, power-aware router and link mechanisms export knobs by which network power and performance can be adjusted. Policies then control these knobs to deliver network power-performance that meet design goals. Self-regulating power-aware networks alleviate designers from the daunting task of reconciling two highly divergent goals - high performance, and low power. In highlighting and demonstrating the importance of power awareness in networks, our proposed research advances a deeper understanding of networks from a fresh perspective, with a potential for high-impact contributions in research. In addition, we see our research leading to power efficiency being an integral part of networking curriculum.