Swarming Scalable Content Delivery for the Masses

Due to the high cost of a Content Distribution Network, most Internet users are not able to scalably deliver content to large audiences. In this paper we study swarming, a scalable and economic content delivery mechanism that combines peer-to-peer networking with parallel download. First, we define a swarming architecture that generalizes the basic delivery mechanism in popular swarming protocols such as Gnutella and BitTorrent. We then conduct a comprehensive performance study of swarming delivery, using a variety of workloads. Our results show that swarming scales with offered load up to several orders of magnitude beyond what a basic web server can manage. Most impressively, swarming enables a web server to gracefully cope with a flash crowd, with minimal effect on client performance. During the course of our study we illustrate the benefits and limitations of a basic swarming protocol and identify several key opportunities for performance improvements.