Optimizing the energy consumption of large-scale applications

The energy consumption of large IT infrastructures is becoming a major concern, since it represents one of the principal operation costs. While modern devices (e.g., processors, disks) have the capability of reducing their power consumption by running at lower speed, this feature must be used with care, as slowing down devices may increase the execution time of the applications beyond acceptable limits. In this paper we propose the qoS AWare energY managER (SAWYER), a framework for dynamically reducing the energy requirement of large-scale applications subject to response time constraints. SAWYER identifies the optimal performance/power consumption tradeoff such that the overall energy requirement is minimized and the application response time is kept below a pre-defined maximum value. This is achieved using a control loop based on a greedy optimization strategy which uses a Queueing Network performance model to quickly evaluate different power settings, ensuring that the expected system response time is kept below the threshold. SAWYER is completely transparent and does not require any modification of the application itself.