Model predictive control for energy-efficient operation of data centers with cold aisle containments

TL;DR

This paper introduces a model predictive control approach to optimize cooling and server operation in data centers with cold aisle containment, significantly reducing energy consumption through convex optimization techniques.

Contribution

It develops a novel MPC framework that jointly optimizes cooling and server activity, transforming the problem into a solvable convex optimization model.

Findings

MPC effectively reduces data center energy use.

Convex optimization enables efficient solution of control problems.

Numerical simulations confirm the approach's practicality.

Abstract

In this paper, we study a problem of controlling cooling facilities and computational equipments for energy-efficient operations of data centers. Although a plethora of approaches have been proposed in previous literatures, there is a lack of rigorous methodologies for effectively addressing the problem. To bridge this gap, we propose MPC frameworks for jointly and optimally tuning the Computer Room Air Conditioner (CRAC) supplying air temperature as well as the number of active servers for minimizing the overall energy consumption. We specifically find that, when a standard model of data centers with contained cold aisles and cooling facilities are used, the optimization problems arising from the MPC framework can be transformed to convex optimization problems that can be solved efficiently. We present several numerical simulations to illustrate the effectiveness of our theoretical results.