Community Based Discovery in Peer to Peer Networks

Existing discovery algorithms for peer to peer networks are based on broadcast of query messages over the relationship connectivity among entities in the network.A query message contains search key words, and it is broadcast with a TTL (time to live) value to limit the scope of search and to reduce high overhead associated with broadcast. Although the use of TTL reduces the amount of overhead, it is difficult to determine an optimal TTL value for a given network, and thus, scalability of existing discovery algorithms may be limited. In addition, in existing discovery algorithms, all search target entities that match the query keywords return a query hit, independent of if they are considered useful in the past by users (i.e., query originator), potentially leading to a large number of non-useful query hits returning to a user. In this paper, we propose a new discovery algorithm for peer to peer networks based on the community and user preference. In the proposed discovery algorithm, each entity gradually acquires keywords of entities that it has a relationship with and represents a community (i.e., a group of entities that it has a relationship with). In addition, users in the propsoed discovery algorithm express their levels of satisfaction with the query hits they receive. In the proposed discovery algorithm, these communities and user satisfaction levels are used in determining where a query message may be forwarded to in search and whether an entity received a query message should return a query hit to the query originator. Community based query message forwarding reduces unnecessary forwarding of query messages, makes the discovery process robust to dynamic changes in the entity relationship connectivity, and eliminates the need for the use of TTL. Query hit returns based on user satisfaction levels reduce the number of unnecessary hits returns. In this paper, we examine characteristics of the proposed discovery algorithm and obtain various performance measures through simualtions. Our simulation results show that the proposed algorithm is scalable and robust to dynamic changes in a network.