games project


Jun 12, 2012

Workload Prediction based Resource Management for Green Data Centers

Adrian Cosinschi, Cristian Botau, Ionut Anghel, Tudor Cioara, Ioan Salomie, Workload Prediction based Resource Management for Green Data Centers, 21st IEEE International Conference on Collaboration Technologies and Infrastructures, Collaborative and Autonomic Green Computing, 2012,

Category: Publications
Posted by: alerox
In this paper we approach the problem of data centers’ high energy consumption by proposing a prediction based controller for optimal and energy efficient allocation of the hardware computing resources to the incoming workload. The controller analyzes the incoming workload virtual tasks’ requests for hardware computing resources and classifies them in predefined tasks classes. For each class, the tasks’ arrival rates are computed and the next period workload is predicted using a neural network. An optimal green data center situation is determined, in which all the predicted virtual tasks are accommodated and the hardware computing resources utilization is minimized. By comparing the current and predicted data center situations, the controller decides and constructs an adaptation action plan to accommodate the incoming workload in an energy efficient manner. The controller implementation is tested on a real test case data center and significant energy savings are reported