PeerTalk

Recent Internet advancement has made large-scale live streaming a reality. Although previous work has studied the feasibility of supporting stream content delivery using peer-to-peer (P2P) architectures, little is known whether it is feasible to provide large-scale multiparty voice-over-IP (MVoIP) services using application end points such as peer hosts. The MVoIP service allows a group of people to freely communicate with each other via the Internet, which can be used in many important applications such as massively multiplayer online gaming telechorus, and online stock trading. Different from conventional conferencing systems that impose explicit or implicit floor controls, it strives to provide a more flexible MVoIP service that allows any participant to speak at anytime. By speaking, it mean not only the uttering of words but also nonverbal activities such as shouting, singing, cheering, and laughing that are common in interactive and spontaneous applications such as online gaming. For example, in the Internet gaming application, MVoIP services allow game players to easily communicate with each other for deploying strategies and game spectators to cheer up players. The emerging collaborative distributed virtual environment applications such as inhabited television and digital virtual world can support large online communities and highly interactive social events where it is common to have overlapping audio transmissions from multiple participants. Using the multicast approach, the system needs to distribute multiple audio streams concurrently from all active speakers to all participants. Although multicast is well suited for broadcast applications that usually involve one active speaker, it becomes inefficient for interactive and spontaneous applications that often include many simultaneous speakers. The system can be overloaded by processing many audio streams concurrently. Moreover, since any participant is allowed to produce audio streams at any time, the system need to maintain a large number of multicast trees for all participants, which can incur a lot maintenance overhead especially in dynamic P2P environments where peers can dynamically leave or join the system. The audio mixing scheme can effectively reduce the number of concurrent streams, which first mixes the audio streams of all active speakers into a single stream and then distribute the mixed stream to all participants. However, centralized audio mixing lacks the scalability desired by P2P applications that often have large groups and many concurrent voiceover-IP (VoIP) sessions. For example, the existing most popular VoIP system Skype can only support conferencing sessions with at most five people. The audio mixing scheme can effectively reduce the number of concurrent streams, which first mixes the audio streams of all active speakers into a single stream and then distribute the mixed stream to all participants. However, centralized audio mixing lacks the scalability desired by P2P applications that often have large groups and many concurrent voiceover-IP (VoIP) sessions. For example, the existing most popular VoIP system Skype can only support conferencing sessions with at most five people.