Capacity Planning and Power Management to Exploit Sustainable Energy Daniel Gmach, Jerry Rolia, Cullen Bash, Yuan Chen, Tom Christian, Amip Shah, Ratnesh Sharma, Zhikui Wang HP Labs Palo Alto, CA, USA e-mail: {firstname.lastname}@hp.com Abstract—This paper describes an approach for designing a power management plan that matches the supply of power with the demand for power in data centers. Power may come from the grid, from local renewable sources, and possibly from energy storage subsystems. The supply of renewable power is often time-varying in a manner that depends on the source that provides the power, the location of power generators, and the weather conditions. The demand for power is mainly determined by the time-varying workloads hosted in the data center and the power management policies implemented by the data center. A case study demonstrates how our approach can be used to design a plan for realistic and complex data center workloads. The study considers a data center’s deployment in two geographic locations with different supplies of power. Our approach offers greater precision than other planning methods that do not take into account time-varying power supply and demand and data center power management policies. Keywords-component; Capacity Planning, Energy Supply Management, Energy Demand Management, Power Capping I. INTRODUCTION Significant research is underway to develop technologies that improve the energy efficiency of data centers and reduce their dependence on the power grid. On the demand side, virtualization technology is being used to consolidate workloads and improve IT utilization [1]; cutting-edge cooling technologies such as dynamic smart cooling [2] and air-side economizers further help improve data center energy efficiency. On the supply side, distributed generation and renewable energy sources are increasingly being deployed. However, the joint behavior of these technologies in an integrated supply─demand context is hard to predict. The goal of this work is to support the design of a power management plan that matches the supply of power with the demand for power in data centers. We define a power management plan as a choice for the peak grid power, a mix of renewable energy sources, energy storage, and data center server power management policies. We assume the data center also has a backup plan that may require the use of more grid power or diesel generators when necessary. A sensitivity analysis using our approach can determine how often such a backup plan is expected to be employed. Although this work focuses on design time, the resulting plan and its policies can be used to optimize the real time management of the data center. Peak grid power usage is often a concern for data centers as it can affect the power infrastructure of grid power providers. Contracts are often heavily influenced by peak usage [3] and can impose penalties for exceeding an agreed upon peak. On the supply side, power may come from a primary power source such as the power grid, from local renewable sources, and from energy storage subsystems. The supply of renewable power is often time-varying in a manner that depends on the source that provides the power, the location of power generators, and the local weather conditions. On the demand side, data center power consumption is mainly determined by the time-varying workloads hosted in the data center and its power management policies. We assume that the data center has pools of servers that execute consolidated workloads, and that these servers support power management policies such as capping the amount of power used by the servers. Power may be capped if the demand for power from the grid exceeds the choice for available peak grid power. We consider two types of power capping policies in this work: server power capping and pool power capping. Server power capping exploits dynamic processor frequency scaling to adjust power used by servers. Pool power capping dynamically varies the number of running servers based on the availability of power and the demand of the workloads. The capping methods are complementary. We describe a trace-driven, simulation based capacity planning tool that is able to simulate power management activities in data centers and estimate the impact of power management plans on application performance and power usage. Traces provide the detailed information needed to: effectively consolidate workloads onto servers; report on metrics that express how often demands are satisfied; and, to estimate the impact of consolidation on workload performance and power consumption. A chosen plan can then be used during the operation of the data center to achieve the desired behavior. A case study involving three months of data for 138 SAP applications in a real data center is used to evaluate the effectiveness of a set of power management plans. Our approach gives greater precision than other methods that do not take into account the time-varying resource usage of the data center, the location specific weather data for renewable power sources, the time-varying supply of power, the power management policies, and the impact of energy storage technologies. The paper is organized as follows. Section II gives an overview of our overall approach. Section III documents the implementation of server and pool power management and data center energy storage in the simulator. Section IV provides characterizations of workload demand variability and renewable power supply variability. The case study is presented in Section V. Related work is described in Section VI. Section VII offers a summary and the concluding remarks along with a description of future research.