Energy-aware resource management of virtual machines under thermal and placement constraints

We describe a new approach to minimize the power used by servers and computer room air conditioners (CRACs) in data centers by controlling them in an integrated way. This approach consists of air conditioning control using a new thermal model to keep inlet temperatures of servers under certain limits and allocation control of virtual machines (VMs) for servers with placement constraints to minimize server and CRAC power consumption. We formulated the resulting optimization problem as a mixed integer quadratic programming problem. We also developed an energy-aware resource management system which is coordinated with a VM management system. We conducted an experiment in our server room and found that our approach resulted in up to 30% less power consumption of CRACs compared to the case with the most inefficient allocation of VMs.