Performance-aware power control of server clusters

One of the key challenges for high density servers is power consumption. High energy consumption causes excessive heat generation which increases system failure and consequently cooling costs. This paper presents a theoretical and experimental framework to control the peak power consumption of servers at the cluster level. We propose an enclosure-level controller that allocates power among servers based on performance needs. Also, we generalize a classic P controller from a Single-Input-Single-Output (SISO) control problem to a Multi-Input-Multi-Output (MIMO) control problem to guarantee that the total power consumption of the cluster stay at or below a power set point. Experimental results show that our controller outperforms a state of the art controller, by having more accurate power control.