Dynamic Energy Cost Conservation for Distributed Edge Clouds Utilizing Online Mini-Batch Learning

Distributed edge clouds (ECs) have been recently shown with remarkable advantages in enhancing customized service provisioning by leveraging user proximity and edge resources. However, operating a massive EC network would inevitably incur a huge amount of energy cost to EC providers, which would offset their operating revenue without proper energy cost management. In this paper, we focus on conserving energy cost of ECs by taking advantage of both electricity price-aware geographical task dispatching and dynamic central processing unit (CPU) provisioning according to the spatiotemporal diversities of electricity prices and user task demands. Due to the significant switching cost of turning CPUs and services on/off, we formulate a multi-timescale energy cost minimization problem that integrates both large-timescale CPU provisioning and service placement, and small-timescale geographical task dispatching and CPU resource allocation. The Lagrange dual decomposition theory is exploited to deal with the spatio-temporal variable couplings. A distributed and online mini-batch learning (MBL) algorithm that relies on parameter approximation for large-timescale decision makings is proposed to learn the optimal Lagrange multipliers. Simulation results show the outstanding performance of the MBL algorithm.