PeerCast: Churn-resilient end system multicast on heterogeneous overlay networks

Abstract

The lack of wide deployment of IP multicast in the Internet has prompted researchers to propose end system multicast or application-level multicast as an alternate approach. However, end system multicast, by its very nature, suffers from several performance limitations, including, high communication overheads due to duplicate data transfers over same physical links, uneven load distribution caused by widely varying resource availabilities at nodes, and highly failure-prone nature of end-hosts. This paper presents a self-configuring, efficient and churn-resilient end-system multicast system called PeerCast. Three unique features distinguish PeerCast from existing approaches to application-level multicasting. First, with the aim of exploiting network proximity of end-system nodes for efficient multicast subscription management and fast information dissemination, we propose a novel Internet-landmark signature technique to cluster the end-hosts of the overlay network. Second, we propose a capacity aware overlay construction technique to balance the multicast workload among heterogeneous end-system nodes. Third, we develop a dynamic passive replication scheme to provide reliable end system multicast services in an inherently dynamic environment of unreliable peers. We also present a set of experiments showing the feasibility and the effectiveness of the proposed mechanisms and techniques.