Green data center design with autonomic power-aware workload allocation -

Abstract

Cloud Computing has revolutionized the technology world and businesses by enabling on-demand provisioning of computing resources allowing users to store, process data and run applications remotely. This entailed the deployment of large-scale data centers containing thousands of computing nodes. However, the energy consumed by Cloud Data Centers today is huge resulting in overwhelming electricity bills and carbon dioxide footprints. In 2010, data centers consumed around 1.5percent of the worldwide electricity and are likely to consume further. This thesis presents a novel approach to reduce energy consumption in a data center. In particular we present a mathematical model that represents the energy dissipation optimization problem. We analytically formulate the server selection problem and the supply air temperature as a Non Linear Programming (NLP), and propose an algorithm to solve it dynamically. A simulation study on SimWare, using real workload traces, shows considerable savings for different data center sizes and utilization rates as compared to three other classic algorithms. The results prove that the proposed algorithm is efficient in handling the energy-performance trade-off. Moreover, they demonstrate that our algorithm provides significant energy savings and maintains a relatively homogenous and stable thermal state at the different rack units in the data center.