Nozzilla: A Novel Peer to Peer Architecture for Video Streaming

Abstract

Many peer-to-peer video streaming systems use application-level multicast where the peers are responsible of forwarding the video packets between them, usually following the path of one or more tree overlays. Since the performance of such systems depends on the uplink capacity available at each peer (measured as bandwidth, delay and jitter), previous work focuses on sharing the load responsibility between all peers by dividing the video traffic into several stripes and requiring a peer to be an interior node in one multicast tree. However, although these techniques consider the uplink capacity limit, they do not adapt easily for situations where the resources are either very abundant or very scarce. This thesis proposes Nozzilla, a new approach based on structured peer-to-peer that dynamically takes into account the available capacity resources. Increased efficiency and reliability is achieved in several ways. First, the peers in the overlay are grouped according to their current resources, improving the search for a suitable parent inside the multicast tree. Second, the system uses path diversity by splitting the video traffic in several stripes and distributes each stripe along a different multicast tree. Third, the protocol can be used in a quality-of-service enabled network, which can assign different classes of service for each video stripe. This makes possible the usage of techniques such as multi-description coding and scalable video coding and is more robust to failures since the loss of a single stripe does not cause the loss of video reception. Finally, the algorithm improves the performance of the multicast tree by increasing the load distribution between peers and reducing the resource demand on the root.