Audio/video streaming applications are becoming increasingly popular over Peer-to-Peer (P2P) networks. Supporting streaming applications over P2P networks is challenging for several reasons including: 1) heterogeneous and asymmetric bandwidth connectivity among participating peers, and 2) lack of support for QoS over today's Internet. In this project, we design and develop a comprehensive framework to support streaming application over P2P networks, called P2P Adaptive Layered Streaming (PALS), that can achieve the above design goals. PALS is receiver-driven, it allows each receiver to independently determine "how to join a session" and "how to deliver the desired stream" from other peers in the session. PALS can support a spectrum of non-interactive streaming application ranging from (lecture-mode) live to playback session.


Projects:

---

A common approach to content delivery in P2P networks is to form an overlay. Most of the existing solution s adopted some variants of application-level multicast, where peers form a single (or multiple), topologically-aware (or delay-aware) tree structure(s) to cooperatively distribute a content. These solutions accommodate scalability by minimizing network load across the tree. However, they are unable to maximize delivered quality to individual peers because of the inherent limitation of delivered bandwidth through a tree-structure overlay of peers with heterogeneous and asymmetric bandwidth. Extending these solutions to multiple tree does not address this fundamental limitation either.

PALS abandons this approach and tackles this problem from a new angle. In PALS each peer is allowed to connect to the overlay at multiple points (or receive content from multiple other peers). More importantly, each peer independently learns about other peers in the overlay, selfishly determines the best way to connect to the overlay in order to maximize its own quality, and dynamically adjusts its connections to the overlay as network or overlay condition change. Therefore, competing peers form an unstructured, bandwidth-aware overlay that maximize delivered quality to each peer. Furthermore, such an overlay can gracefully accommodate dynamics of peer participation. We are currently developing this overlay construction strategy, and will post further details in a near future.

One key issue is to simultaneously stream a single video/audio content from multiple senders. This is challenging because all the pair-wise connections between peers must be congestion controlled, and thus the available bandwidth from each sender has potentially wide variations during delivery. This implies that the delivery mechanism should be quality adaptive. To accommodate bandwidth heterogeneity, we assume that content has layered structure (e.g. either layer or multiple description encoded streams).

To seamlessly support asynchronous peers who join a session with different delays, PALS incorporates a distributed caching mechanism. Each peer contributes an arbitrary amount of cache space, and the collection of all participating peers form a distributed and heterogeneous cache space. Each stream should be partially cached at different peers in order to enable multi-sender delivery while achieving high caching efficiency for the distributed cache space. The key challenge is how to manage this distributed cache space efficiently in order to maximize both delivered quality to asynchronous clients and efficiency of distributed cache space.

Acknowledgment and Disclaimer