Energy proportional data center applications

Data center applications pose challenges with respect to their performance and energy requirements. These applications are mostly network oriented applications that are executed in data centers. The energy usage of a data center becomes more important as it scales and, thus, it approaches its physical limitations. The expectation of achieving maximum performance, in terms of latency and throughput, conflicts with those strict energy requirements imposed by the data center itself. The compromise in use today by data center operators that maximizes performance and satisfies respective service level agreements requires that servers remain underutilized, which are able to provide minimal latency and high throughput. On the other hand, servers are not energy proportional which means that an underutilized server consumes more power per unit of performance than a fully utilized one. In this write-up, I am going to present two methods that increase the performance of network applications and one method that aims to balance the QoS of applications and energy consumption of servers in a data center. Finally, I will present my research proposal that aspires to combine available methods into providing a framework for energy proportional data center applications.