Since the electricity bill of a data center constitutes a significant portion of its overall operational costs, reducing this has become important. We investigate cost reduction opportunities that arise by the use of uninterrupted power supply (UPS) units as energy storage devices. This represents a deviation from the usual use of these devices as mere transitional fail-over mechanisms between utility and captive sources such as diesel generators. We consider the problem of opportunistically using these devices to reduce the time average electric utility bill in a data center. Using the technique of Lyapunov optimization, we develop an online control algorithm that can optimally exploit these devices to minimize the time average cost. This algorithm operates without any knowledge of the statistics of the workload or electricity cost processes, making it attractive in the presence of workload and pricing uncertainties. An interesting feature of our algorithm is that its deviation from optimality reduces as the storage capacity is increased. Our work opens up a new area in data center power management.
Emerging energy-aware initiatives (such as billing of power usage based on de-coupling between electricity sales and utility profits/fixed-cost recovery) render current capacity planning practices based on heavy over-provisioning of power infrastructure unprofitable for data centers. We explore a combination of statistical multiplexing techniques (including controlled under-provisioning and overbooking) to improve the utilization of the power hierarchy in a data center. Our techniques are built upon a measurement-driven profiling and prediction technique to characterize key statistical properties of the power needs of hosted workloads and their aggregates. As a representative result from our evaluation on a prototype data center, by accurately identifying the worst-case needs of hosted workloads, our technique is able to safely operate 2.5 times as many servers running copies of the e-commerce benchmark TPC-W as allowed by the prevalent practice of using face-plate ratings. Exploiting statistical multiplexing among the power usage of these servers along with controlled under-provisioning by 10% based on tails of power profiles offers a further gain of 100% over face-plate provisioning. Reactive techniques implemented in the Xen VMM running on our servers dynamically modulate CPU DVFS-states to ensure power draw below safe limits despite aggressive provisioning. Finally, information captured in our profiles also provides ways of controlling application performance degradation despite the above under-provisioning: the 95th percentile of TPC-W session response time only grew from 1.59 sec to 1.78 sec.