J.kDenzinger T.kMatsui D.kGrosu K.kHirayama M.kYokoo M.kZanker 1 2 5 3 6 4 4 4

In this study, we introduce a new peer-to-peer (P2P) approach to instant messaging systems based on a fully decentralized network, and where each human owning a peer can control the traffic supported by her system. The control may be based on criteria such as: (a) her desire to help the endpoints of the communication, e.g., based on friendship, (b) her desire to help a cause, e.g., based on the content/topic of the communication, (c) reputation, or (d) the utility brought to her by the handled data. Peers need help to communicate when they are behind NATs. Unlike the P2P chat system used in Skype, no centralized servers are involved (no central redirect server, login server or web server). Providing intrinsic motivation such that peers help with traffic is important in order to eventually make an open source P2P chat system viable. In current non-incentive P2P systems like Skype, availability of open source versions can potentially starve the system of supernodes (since users can disable the resource consuming supernode-function). In our approach, each peer has equal privileges to any other peer. Nodes register their address with other peers of their choice, which can then act as directories on their behalf. Each peer with sufficient resources can voluntarily play the role of directory for her friends or of forwarding incoming messages to peers behind NATs. This paper analyzes key functions of the solution such as incentive management, NAT and firewall traversal, connection establishment and message transfer under different network setups. Introduction Instant messaging has been one of the most successful applications on the Internet, starting with talk and IRC to Twitter and Facebook. One of the most currently used such systems, Skype, is a peer-to-peer (P2P) application. The P2P architecture offers Skype significant robustness and scalability, as well as efficiency, particularly from the perspective of a low latency. Besides depending on a centralized set of login servers that could be used to eventually bring down the whole system, Skype is a closed source software, and it is known to use computational resources of unsuspecting users. While efforts to build open source clients have failed due to secret changes in protocols (Baset & Schulzrinne 2004; 2006; Guha, Daswani, & Jain 2006), it is questionable whether they could in fact survive when the protocol would be open. The issue is that, with open source software, the users can get versions that disable expensive super-node functions that exploit the resources of the nodes. This missing property is called faithfulness (Shneidman, Parkes, & Massoulié 2004). In the closed source chat systems, the only incentive to act as supernode is the fact that binaries do not allow users to disable that function selectively. Since current protocols have no intrinsic incentives to voluntarily act as a supernode, availability of an open source agent can starve the network of super-nodes and lead to the demise of the service. We propose an approach to P2P supernodes that can simultaneously offer advantages similar to Skype, while being fully decentralized, open source, allowing customization and control of one’s resources. Users may thereby enforce that their resources are used only for causes that they want to support. Providing intrinsic incentives for peers to help with traffic flow in P2P network applications is important in order to overcome the most common problems with the open-source P2P paradigm: Free-riding and tragedy of the commons (Ma et al. 2003). Incentives are intrinsic to an application if the participant loses by not performing the actions expected of her due to those actions, and not due to benefits provided separately (e.g., bundled in a pre-compiled software). Incentives To encourage peers to support the chat communication of others, the proposed protocol enables incentives, such as: • Ability to control the traffic passing by her system: – helping the endpoints of the communication (e.g., based on friendship or subject). – rejecting the endpoints of the communication (e.g., based on disagreement). • Getting information concerning who talks to whom and what are they talking about (for research, marketing data or pure curiosity). • Supporting open source P2P chat systems, thereby supporting the freedom of customization. • Opportunity to insert advertisements into the stream. • Ability to offer paid services for supporting the communication. The characteristics of our P2P approach is: network address translation (NAT) piercing ability, democracy, no outside control, full decentralization. In the next section we discuss differences with related work. After introducing the main concepts, section Protocol describes in detail the mechanism used by peers for incentive chat.