A simulator to assess energy-saving techniques in content distribution networks

The scalable and bandwidth-efficient delivery of booming linear and on-demand video services to an increasingly diverse set of screens (laptop, tablet, smartphone, etc.) requires the deployment of telco content distribution networks (CDNs). These CDNs are composed of numerous cache servers located in the telco's data centers at the edge of the aggregation network, close to the end user. The additional cache servers, which are packed with disks to support video-on-demand, need to be designed for energy efficiency to limit the increase of data-center energy consumption, which already accounts for up to half of the data center's total cost of ownership. In this paper, we analyze real HTTP adaptive-streaming traces from an operational CDN to identify workload characteristics that can be exploited to conserve energy. In addition, we present a trace-driven simulator that models the energy consumption of such a CDN down to the level of the cache-server disk, to evaluate potential energy-saving techniques. The traces reveal cyclic load fluctuations that can be exploited in combination with the segregation of original and redundant data inherent in CDNs to save energy by varying the number of powered cache servers and disks according to the load.