The enormous power consumption of cloud data centers poses serious financial and environmental concerns. Server consolidation, server throttling, and power capping emerge as several of the numerous approaches proposed to enhance the power efficiency of data centers. While these methods notably increase data center power efficiency by reducing the power consumption of data center servers, a side effect emerges in which the performance of hosted applications can be negatively impacted, leading to user complaints. Thus, data center operators must grasp and navigate power-performance tradeoffs. The motivation behind the study conducted in this thesis is to formulate strategies for managing power-performance tradeoffs in cloud data centers. To such an end, the connections between the power usage of cloud data center servers and the performance of applications hosted on these servers are investigated, models that capture these connections are designed, and controllers aimed at diminishing the power usage of data center servers while considering application performance are developed.
In this thesis, I make the following contributions. First, I summarize taxonomies for various power capping methods, elastic resource scaling techniques, and application performance metrics. Second, I address the question of whether we can automate the process of power-performance controller generation for latency-sensitive web applications, ensuring that the generated controllers reduce power allocation while meeting SLOs and offer equal or improved performance compared to state-of-the-art techniques. This is achieved by designing DDPC, a system for autonomous data-driven controller generation for power-latency management. Third, I introduce PoVerScaler, a framework for effectively coordinating elastic resource provisioning techniques and power capping methods under performance constraints. Finally, I investigate the effect of fractionally scaled CPU cores on the performance of latency-sensitive applications and server power consumption as well as propose a controller that uses fractional core scaling as an actuator for managing power-performance tradeoffs.