A distributed energy saving approach for Ethernet switches in data centers

Abstract

With popularity of data centers, energy efficiency of Ethernet switches in them is becoming a critical issue. Most existing energy saving approaches use a centralized methodology that assumes global knowledge of data center networks. Though these approaches can achieve nearly optimal energy saving for static traffic patterns, they are not suitable when the traffic patterns can change rapidly or the data centers have a large size. To overcome these limitations, this paper proposes a novel distributed approach called eAware that dynamically idles a port or a switch to save energy by examining the queue lengths and utilizations at switch ports. Through extensive simulations in ns-2, we compare eAware with an existing energy oblivious approach, showing that eAware can save 30%-50% on the total energy consumption by switches in data centers, and only increases the average end-to-end delay of packets by 3%-20% and the packet loss ratio by 0%-0.9%.