Abstract:
To leverage the elastic nature of cloud computing, a solution provider must be able to accurately gauge demand for its offering. For applications that involve swarm-to-cloud interactions, gauging such demand is not straightforward. In this paper, we propose a general framework, analyze a mathematical model, and present a prototype implementation of a canonical swarm-to-cloud application, namely peer-assisted content delivery. Our system – called Cyclops – dynamically adjusts the off-cloud bandwidth consumed by content servers (which represents the bulk of the provider's cost) to feed a set of swarming clients, based on a feedback signal that gauges the real-time health of the swarm. Our extensive evaluation of Cyclops in a variety of settings – including controlled PlanetLab and live Internet experiments involving thousands of users – show significant reduction in content distribution costs (by as much as two orders of magnitude) when compared to non-feedback-based swarming solutions, with minor impact on content delivery times.
