Energy Efficiency in Content Delivery Network

NiemahIzzeldin Mohamed Osman

Department of Computer Science, College ofComputer Science and Information Technology, Sudan University of Science and Technology,Sudan.

======

The increasing popularity of bandwidth-intensive video Internet services has positioned Content Distribution Networks (CDNs) in the limelight as the emerging provider platforms for video delivery. The goal of CDNs is to maximise the availability of content in the network while maintaining the quality of experience expected by users. This is a challenging task due to the scattered nature of video content sources and destinations. Furthermore, the high energy consumption associated with content distribution calls for developing energy-efficient solutions able to cater for the future Internet. This thesis addresses the problem of content placement and update while considering energy consumption in CDNs.

First, this work contributed a new energy-efficient caching scheme that stores the most popular content at the edge of the core network and optimises the size of cached content to minimise energy usage. It takes into account the trend of daily traffic and recommends putting inactive segments of caches in sleep-mode during off-peak hours. Our results showed that power minimisation is achieved by deploying switch-off capable caches, and the trend of active cache segments over the time of day follows the trend of traffic.

Second, the study explores different content popularity distributions and determines their influence on power consumption. The distribution of content popularity dictates the resultant cache hit ratio achieved by storing a certain number of videos. Therefore, it directly influences the power consumption of the cache. The evaluation results indicated that under video services where the popularity of content is very diverse, the optimum solution is to store the few most popular videos in caches. In contrast, when video popularities are similar, the most power efficient scheme is either to cache the whole library or to avoid caching completely depending on the size of the video library.

Third, this thesis contributed an evaluation of the power consumption of the network under real world TV data and considering standard and high definition TV programmes. We proposed a cache replacement algorithm based on the predictable nature of TV viewings. The time-driven proactive cache replacement algorithm replaces cache contents several times a day to minimise power consumption. The algorithm achieves major power savings on top of the power reductions introduced by caching.

CDNs are expected to continue to be the backbone for Internet video applications. This work has shown that storing the right amount of popular videos in core caches reduces from 42% to 72% of network power consumption considering a range of content popularity distributions. Maintaining up-to-date cache contents reduces up to 48% and 86% of power consumption considering fixed and sleep-mode capable caches, respectively. Reducing the energy consumption of CDNs provides a valuable contribution for future green video delivery.