Thermal-aware load provisioning for server clusters by using model predictive control

Energy efficiency and cooling availability of data centers have drawn considerable attentions from academia and industry. This paper considers load provisioning for a cluster of servers with different inlet temperatures. A model predictive control approach is proposed to optimize the on/off settings and the load distributions of servers. This approach minimizes the energy consumption while limiting outlet air temperatures of servers within a specified range. Web request traces of the FIFA World Cup 98 are used as workloads in the simulation. Results indicate that the proposed control strategy achieves significant saving of server energy and high efficacy of temperature control.