Evolutionary P2P Networking for Realizing Adaptive Networks

Abstract

Recently, a peer-to-peer (P2P) network model became very popular. This model is different from a conventional client-server network model. While a conventional client-server network model explicitly distinguishes nodes providing services (servers) from nodes receiving services (clients), a P2P network model does not assign fixed roles to nodes. One type of P2P networks are unstructured P2P networks that do not include any mechanism to manage data locations and which consist only of nodes that communicate with each other through direct connections. A direct connection between two nodes in unstructured P2P networks is represented by a logical network link, and therefore, a structure formed by logical links and nodes, that is, a P2P network topology, can be formed freely. Thus, unstructured P2P networks are flexible. However, they require some mechanisms to realize quick, accurate, and reliable searches. A free-form P2P network topology can be a control object for realizing such searches. One method that can adaptively modify a free-form topology of a running unstructured P2P network for quick, accurate, and reliable searches is an evolutionary algorithms (EA) inspired by biological genetics and evolution. Although EA is not the only way for adjusting the entire network topology of a running P2P network, EA would be suitable for this modification because EA can hold several solution candidates (i.e. future options for network topologies) simultaneously in a running P2P network. An evolutionary P2P networking technique (EP2P) and a parallel evolutionary P2P networking technique (P-EP2P) are both a fusion technique of an EA and an unstructured P2P network, and optimize the manner in which the nodes belong to different P2P network topologies in realtime. We will introduce EP2P and P-EP2P and show simulation results that validate both techniques.