Performance analysis of parameters affecting power efficiency in networks

Abstract

Networking devices in ISP networks and data centers have been deployed in an over-provisioning and redundant manner to meet the worst case traffic loads and to quickly recover from failures. These devices are idle or semi-idle most of the time with full power consumption. Today, there is widespread interest in techniques that help in reducing energy waste and achieving high levels of energy-efficiency. Solutions proposed in literature are either topology-oriented (ESTOP) or traffic-oriented (ESTA). In this paper, we propose a Hybrid Heuristic for Green Networking (HHGN) that exploits the parameters considered in both ESTOP and ESTA to identify a subset of network elements that can be switched off such that the power saving is maximized under network connectivity and edge bandwidth utilization constraints. We introduce two new heuristics, w-BFLP (weighted Betweenness, Flow, Links, Power) and w-BFP (weighted Betweenness, Flow, Power), that sort the nodes and the edges respectively in increasing order of their importance in the topology. Nodes and edges are then switched off from the least important until the connectivity threshold or maximum bandwidth utilization constraint on the edges as specified by the user is reached. We compare our approach with ESTOP and ESTA for edge optimization in terms of power gain, mean utilization of edge bandwidth, percentage of sleeping edges, fairness index, trade-off between power gain and edge utilization, and increase in the length of re-computed paths for real ISP and FatTree topologies using different power models. Experimental results show that HHGN gives the best performance independent of power models, traffic matrices and topologies tested.