Energy-Aware Automatic Optimization of Resource Allocation in CloudSim
Patricia Arroba. Universidad Politécnica de Madrid
Host Institution: The Cloud Computing and Distributed Systems (CLOUDS) Laboratory at The University of Melbourne
Computational demand on data centers is increasing due to growing popularity of Cloud applications. Nowadays, this industry consumes about 2% of the worldwide energy production and the carbon footprint generated by cooling systems is expected to overtake airline industry emissions by 2020.
Furthermore, as Cloud applications expect services to be delivered as per Service Level Agreement (SLA), power consumption in data centers has to be minimized while meeting this requirement whenever it is feasible. Also, Cloud workloads vary significantly over time, making optimal allocation and DVFS configuration not a trivial task. Thus, Cloud providers need to implement an energy-efficient management of physical resources to meet the growing demand of their services while ensuring Quality of Service (QoS). Understanding the relationship between power, cooling, DVFS and workload consolidation is crucial to enable energy-efficient management at the data center level.
My research, at Universidad Politécnica de Madrid (UPM) in tight collaboration with Universidad Complutense de Madrid (UCM), proposes new trade-offs between energy, workload consolidation and performance that help on combining DVFS with power and thermal-aware strategies. In this scope, the key contribution of my internship at CLOUDS Lab focuses on addressing the energy challenge in Cloud data centers from a proactive perspective. We propose a consolidation policy that jointly minimizes consumption while maintaining QoS. We have performed an extensive evaluation on the CloudSim toolkit, a state-of-the-art simulator developed at CLOUDS Lab, using real Cloud traces and an accurate power model based on data gathered from real servers. Our results demonstrate that combining DVFS awareness with workload management provides substantial energy savings around 37.86% for scenarios under dynamic workload conditions.
I would like to thank HiPEAC for giving me the excellent opportunity to collaborate with a group that is clearly an international reference in Cloud computing optimizations. This internship has allowed us to start a relationship between UPM, UCM and CLOUDS Lab to perform future research in collaboration. I also need to thank members of the CLOUDS Lab at The University of Melbourne for their comments and support, especially Prof. Rajkumar Buyya for his supervision and guidance.