New Performance Models for Large-Scale Content Delivery Networks
03 Jun 2014
Large-scale Content Delivery Networks (CDNs) are a key component of today's information infrastructure. To enable future disciplined engineering of such infrastructure it is critical to have a fundamental understanding of engineering tradeoffs amongst user percieved download performance, reliability and recovery costs, and energy consumption.
WNCG Ph.D. student Virag Shah and Prof. Gustavo de Veciana devised a new class of models and analysis techniques that enable the performance analysis of CNDs where files are replicated and available for parallel download from multiple servers and peers. Their analysis shows that file transfer delays are inversely proportional to the number of replicas, but perhaps surprisingly depend only logarithmically on the server loads.
The WNCG team's paper is a first attempt to evaluate performance, reliability, costs, and energy tradeoffs for large-scale content delivery networks.
Presentation slides: Infocom 2014