StrandCast: Peer-to-peer content distribution for latency tolerant applications

Abstract

Existing streaming peer-to-peer distribution architectures are designed to balance inherent trade-offs between stretch, stress, and robustness to churn. Several rely on multiple description coding in order to facilitate mesh-based or forest-based approaches. We believe, however, that a number of important applications such as broadcast video, pyramid video-on-demand, digital fountains, and data cycles to name a few exhibit a great deal of latency tolerance. For these applications, stretch is immaterial as long as join/leave latencies are minimized. If stretch is no longer a concern, simply chaining peers into a line emerges as a natural design choice for minimizing stress. The realities of loss, churn, and heterogeneous peer capabilities make implementing such a design challenging. In this paper, we show that by using a layered data model, a linear distribution scheme can be very effectively achieved and can be superior to MDC-based schemes. The paper describes our system called StrandCast and presents simulation results demonstrating its effectiveness and comparing it to SplitStream.