Rate Adaptation for Energy Efficiency in Packet Networks

An ideal model for energy consumption in a packet network is one where energy use is proportional to traffic forwarding rates [2]. More specifically, energy use in a network should scale with the overall traffic load and should become negligible in network elements that do not process packets. It is well understood that over long timescales (hours or days) network links can become idle for extended periods of time, and that powering them off during those extended periods can yield significant energy savings [5]. The goal of this paper is to determine if it is feasible to extend the application of such energy saving techniques to much shorter timescales (milliseconds or seconds) and achieve energy proportionality without sacrificing network performance in terms of packet losses and queueing delays. Two promising methods for reducing energy consumption at short timescales are rate scaling (RS) and sleep-state exploitation (SSE) [8], [10], [15]. With rate scaling, the clock frequency of a data-path device changes over time to match the traffic processing rate of the device with the input traffic load. By also scaling the voltage of the power supply with the clock frequency, as allowed by dynamic voltage and frequency scaling (DVFS) technologies [6], [7], [12], a rate-scaling device can rely on a cubic (and therefore convex) relationship between power and traffic processing rate to reduce its overall energy consumption. With sleep-state exploitation, the network device alternates between a full-capacity state where it operates at maximum clock frequency as long as traffic is available for processing, and a sleep state where most functions are disabled and power consumption is much lower.