Performance evaluation of topology aware super peer selection methods in ALM networks

One-to-many video streaming service using Application Level Multicast (ALM) has been receiving attention recently, as seen in the popularity of P2P-TV software. With the growth of these services, the problems of bandwidth inefficiency of P2P logical link has emerged due to unawareness of underlying physical topology in multicast tree construction. To solve this problem, the approach to incorporate super peers as “always-on” parent peers in the multicast tree has been widely studied. However, most of existing studies in this approach focus proximity aware routing assuming that super peer locations are predetermined. With the advance of user PC specification and broadband access environments, participating peers qualified for super peer candidates are expected to substantially increase. Adequately selected super peers among them considering underlying physical network topology can considerably improve bandwidth efficiency and reduce delay. This paper presents performance comparison of super peer selection methods for ALM tree construction which aims at improving efficiency by taking account of physical network topology information. The results show that these methods work more effectively in networks whose topologies resemble a random network compared with those similar to a scale-free network. Further, adaptive selection of super peers according to peer density distribution in the network can be also effective for some peer distribution patterns of the network.