Green networking strategies for distributed computing

The high bandwidth and speed of wavelength division multiplexing (WDM) optical networks are allowing efficient use of distributed computing and storage resources that would not have been feasible earlier. There has been much focus, in recent years, on the development of “green” techniques that reduce the energy consumption of the computing and storage facilities at the network nodes. However, a significant amount of energy overhead is also incurred in the process of transmitting large amounts of data over the network. This includes energy consumption of various network components such as routers, switches, in-line amplifiers and transponders. In this paper, we focus on energy minimization techniques for high-speed communication over an optical network. Given a set of communication requests, along with the bandwidth needs, and specified start and end times for each request, we propose a new approach that exploits knowledge of demand holding times to intelligently share resources among non-overlapping demands and reduce the overall power consumption of the network. We first present a simple shortest path heuristic that significantly outperforms traditional holding time unaware (HTU) approaches. We then present a novel Genetic Algorithm (GA) based strategy that jointly minimizes both power consumption and transceiver cost for the network and achieves additional energy savings, even compared to our first approach.