Smart Energy Conservation in Data Centers Using Machine Learning Based Software-Defined Networking

Abstract

The rapid growth of Data Centers (DC) poses the problem of heavy energy consumption. The servers consume most of the energy in the DC, but the network infrastructure also uses a great deal of energy. It is therefore necessary to minimize the energy consumption of DC. DC Energy Conservation Strategy has various directorates, including energy-efficient methods for servers and networks, blended for servers and networks, cooling and renewable energy. As the literature revealed, many existing ways are available, but there are many problems with the current strategies, such as the issues include loss of packets, systems with rigid approaches with low modifications, obsolete methodologies for newer hardware devices, methods tested on small networks, and the amount of energy saved compared to the risk and financial cost involved is not worth it. This research eliminates the risk of packet loss up to some extent by securing as well improving the network efficiency. The study aims to construct a way to save energy in DC by employing the software-defined networking technique and modifying the existing Elastic Tree to maximize power savings, minimize the performance effect, and guarantee fault tolerance. It can consider improving the existing optimizer in elastic tree methodology and introducing new factors like power control based on traffic pattern analysis. A Mininet emulator is utilized to develop a virtual DC network. An optimizer is designed using different models and includes a floodlight controller that monitors traffic flow and creates a custom network topology by tracking the traffic pattern in DC. This custom network topology keeps the network's essential components active and power down the unneeded parts, including many links and switches, to meet the best performance. The custom topology suggested demonstrating in J-Graph. This exploration further leads to optimizer analysis, a step toward improving smart energy conservation.