Abstract: We discuss a number of message-passing algorithms that can be efficiently used to search through power-law networks. Most of these algorithms are meant to be improvements for peer-to-peer file sharing systems, and some may also shed some light on how unstructured social networks with certain topologies might function relatively efficiently with local information. Like the networks that they are designed for, these algorithms are completely decentralized, and they exploit the power-law link distribution in the node degree. We demonstrate that some of these search algorithms can work well on real Gnutella networks, scale sub-linearly with the number of nodes, and may help reduce the network search traffic that tends to cripple such networks.

To appear in: 'Handbook of Graphs and Networks: From the Genome to the Internet', S. Bornholdt and H.G. Schuster (eds.), Wiley-VCH, Berlin, 2002.

Local Search in Unstructured Networks [PDF]