Peer-to-Peer Receiver-driven Mesh-based Streaming

Peer-to-Peer (P2P) streaming has become an increasing popular approach for one-to-many multimedia streaming applications , mainly because it does not require any special support (e.g., IP multicast or any content distribution infrastructure) from the network. A common theme in P2P streaming systems is that participating peers form an overlay where each peer receives content from one (or multiple) parent peer(s) in a session. Any P2P streaming system consists of two distinct but related components: (i) an Overlay Construction mechanism that organizes participating peers into an overlay structure with certain properties, and (ii) a Content Delivery mechanism that determines how the multimedia content should be streamed to each participating peer through the overlay. The design of these two components should accommodate the following two potentially conflicting goals: (i) Session-level (Global) Goal: The overall network load associated with content delivery should be minimized; (ii) Peer-Level (Local) Goal: The quality of the stream delivered to individual peers should be maximized within their fair share of available resources (mainly bandwidth). The session-level goal should be cooperatively pursued by all peers whereas each peer should separately maximize its own local goal. Note that the peer-level goal could easily be in conflict with the session-level goal. For example, selecting a parent peer that minimizes the overall network load may not maximize the delivered bandwidth to a receiver peer. A few existing approaches to P2P streaming have adopted the idea of " application-level multicast " where participating peers often form a topologically-aware tree structure over which multimedia content is simply " pushed " from the source towards all peers. In essence, the primary (and often the only) goal in this class of solutions is to accommodate the session level goal by constructing an overlay with certain properties. This class of solutions is unable to maximize the quality delivered to individual peers for the following reasons: First, available bandwidth to each peer in a tree-structure overlay is inherently limited. This problem is further aggravated by the following issues: (i) heterogeneity and asymmetry of access link bandwidth among participating peers, (ii) dynamic peer participation (i.e., churn), and (iii) the competition among participating peers for available band-width from parent peers. We note that organizing peers into multiple, diverse trees (e.g., CoopNet [2]) improves re-siliency to churn but may not improve delivered bandwidth to individual peers due to shared bottlenecks among trees. Second, existing solutions do not leverage the available flexibility …