Energy Efficient Content Delivery Networks

10 Jun
Monday, 06/10/2013 9:00am to 11:00am
Ph.D. Dissertation Proposal Defense

Vimal Mathew

Computer Science Building, Room 151

Advisors: Prashant Shenoy & Ramesh Sitaraman

The energy consumed by internet-scale distributed systems contribute significantly to their operating cost. We focus on content delivery networks (CDNs) and demonstrate techniques to save energy while meeting service level agreements (SLAs) and minimizing impact on hardware reliability.

We propose server shutdown algorithms that consolidate traffic within a cluster of servers to achieve over 50% savings while maintaining high availability. We propose a novel technique called cluster shutdown that consolidates traffic between clusters of servers to switch off entire clusters, without increasing operating costs, while conforming to client-perceived SLAs. We show that since cluster shutdown can save both server and cooling energy, it outperforms server shutdown in a range of operating regimes. Further, it has intrinsic architectural advantages in the CDN context. As part of future work, we propose to study non-real-time CDN workloads that can tolerate service delays and explore if such delays can be exploited for additional energy savings.