Optimization of Data and Energy Migrations in Mini Data Centers for Carbon-Neutral Computing

Abstract

Due to large-scale applications and services, cloud computing infrastructures are experiencing an ever-increasing demand for computing resources. At the same time, the overall power consumption of data centers has been rising beyond 1&#x0025; of worldwide electricity consumption. The usage of renewable energy in data centers contributes to decreasing their carbon footprint and overall electricity costs. Several green-energy-aware resource allocation approaches have been studied recently. None of them takes advantage of the joint migration of <i>jobs</i> and <i>energy</i> in green data centers to increase energy efficiency. This paper presents an optimization approach for energy-efficient resource allocation in mini data centers. The observed momentum around edge computing makes the design of geographically distributed mini data centers highly desirable. Our solution exploits both virtual machines (VMs) and energy migrations between green compute nodes in mini data centers. These nodes have energy harvesting, storage, and transport capabilities. They enable the migration of VMs and energy across different nodes. Compared to VM allocation alone, joint-optimization of VM and energy allocation reduces utility electricity consumption by up to 22&#x0025;. This reduction can reach up to 28.5&#x0025; for the same system when integrating less energy-efficient servers. The gains are demonstrated using simulation and a Mixed Integer Linear Programming formulation for the resource allocation problem. Furthermore, we show how our solution contributes to sustaining the energy consumption of old-generation and less efficient servers in mini data centers.