Towards Sustainable Edge Computing through Renewable Energy Resources and Online, Distributed and Predictive Scheduling

In this work, we tackle the energy consumption problem of edge computing technology looking at two key aspects: (i) reducing the energy burden of modern edge computing facilities to the power grid and (ii) distributing the user-generated computing load within the edge while meeting computing deadlines and achieving network level benefits (server load balancing vs consolidation and reduction of transmission costs). In the considered setup, edge servers are co-located with the base stations of a mobile network. Renewable energy sources are available to power base stations and servers, and users generate workload that is to be processed within certain deadlines. We propose a predictive, online and distributed algorithm for the scheduling of computing jobs that attains objectives (i) and (ii). The algorithm achieves fast convergence, leading to an energy efficient use of edge computing facilities, and obtains in the best case a reduction of 50% in the amount of renewable energy that is sold to the power grid by heuristic policies and that is, in turn, used at the network edge for processing.