On adaptive multi-objective optimization for greener wired networks

Abstract

In this study, we develop a novel adaptive Multi-Objective Optimization (MOO) algorithm to jointly minimize the Power Consumption (PC) and Maximum Link Utilization (MLU) of wired computer networks. This novel algorithm, based on Non-dominated Sorting Genetic Algorithm II (NSGA-II), is able to discover the optimal link weights configuration for wired networks based on live network traffic data. The goal is to minimize both PC and MLU when this optimal link weights configuration is applied to the network. The impact of the Internet and wired networks’ growing power consumption on costs and environment continues to be a major concern of network operators. Our adaptive MOO scheme can be applied to any wired network to reduce its power consumption with acceptable load balancing to help achieve greener energy-efficient networks. In order to validate our solution, we run experiments on two different network topologies with various network traffic data to compare the performance of three different methods of updating the link weights: random, delta-weight, and hybrid. The experiment results show that all three methods can find optimal solutions to achieve both reduced PC and MLU. Compared to the default link weights configuration, the optimal solution found by hybrid approach is the best since it can reduce the PC by 35.24%, while reducing MLU by 42.86% for specific traffic pattern on Abilene network topology.