Energy-efficiency evaluation of computation offloading in personal computing

Abstract

Cloud computing has become common practice for a wide variety of user communities. Yet, the energy efficiency and end-to-end performance benefits of cloud computing are not fully understood. Here, we focus specifically on the trade-off between local power saving and increased execution time when work is offloaded from a user’s PC to a cloud environment. We have set up a 14-node private cloud and have executed a variety of applications with different processing demands. We have measured the energy cost at the level of the individual user’s PC, at the level of the cloud, as well as at the two combined, contrasted to the execution time for each application when running on the PC and when running on the cloud. Our results indicate that the tradeoff between energy cost and performance differs considerably between applications of different types. In most cases investigated, the total increase in energy consumption, incurred by running that additional application, was reduced significantly. This shows that research on using cloud computing as a means to reduce the overall carbon footprint of IT is warranted. Of course, the energy gains were more pronounced for energy-selfish users, who are only interested in reducing their own carbon footprint, but these savings came at the expense of performance, with execution time increase ranging from 1% to 84% for different applications.