AHLSS: A Hierarchical, Adaptive, Extendable P2P Live Streaming System

Abstract

With the development of P2P streaming technology, various live streaming systems have been designed. Some measurement studies recently have shown that the most important issue exposed by previous systems is how to balance the benefits between the internet service provider, the internet content provider, and P2P users. This issue is induced by little consideration on management and control. Other problems include how to resolve peers churn and make full use of network bandwidth and apply incentive mechanism among peers. In this work we introduce AHLSS, a P2P system for live streaming which is hierarchical, adaptive, and extendable. We present the architecture of AHLSS which is divided into three layers. The top layer named AP (Administrator Peer) overlay consisted of some authentic servers, which are arranged by ICP (Internet Content Provider) to provide enrollment and authentication for peers to join P2P overlay and start to watch a media channel. Every AP locates in an alliance alone. The middle layer is SP overlay. AHLSS divides peers that participate in data transmission into two classes, including SP and NP, which base on the capability of peers. SP is elected by AP. SP has responsibility to route media packets quickly and reply peers' channel request within the cluster. The under layer consists of many clusters. A cluster is formed by uniting peers which have poor capability and little stability near SP. SP is named cluster-head with the responsibility for management of members in the cluster. Our system's innovative features are the following: It provides the function of enrollment and authentication; AP look up peer's EnrollInfo (a list, which keeps the register information of peers) periodically and apply a decision-tree approach to elect SP. It resolves problems which are associated with the dynamics and heterogeneity of peers; to cope with the problem of topology-mismatch, we design Alliance and Cluster to partition the physical location of peers and make use of the Two-hop Detection Algorithm which belongs to LTM technology to optimize the connection of SP among alliances; different transmission modes are designed on SP overlay and within cluster. Native mesh-pull is adopted in SP overlay and the hybrid mode includes tree-push and mesh-pull which are applied in NP (Normal Peer) overlay. It reduces the delay in chunks transmitted through SP to NP and makes full use of the upload bandwidth of peers; Approach of K degree redundancy cluster-head is adopted by NP, it is used to avoid the inorganized status for NPs which resulted from SP's failure. In this paper, we describe in detail the main algorithms that stand behind the design of AHLSS and present the analysis of architecture for various aspects including adaptability, scalability, efficiency, and robustness.