<PlanetP: using gossiping to build content Addressable peer-to-per information sharing communities>
Replicate a Bloom Filter summary of the entire index at every peer using gossiping. The index contains the names and addresses of all current members and Bloom filters summarizes the set of terms contained in the documents being shared by that member.
Gossiping algorithm: a node x learns a change, every T second x randomly choose a target peer y to tell y of the change. If y has not heard of it, it record the new info and spread the rumor. Every several round of the rumor, x send an anti-entropy message asks the target y to send a summary of its entire directory to x. ( or y piggybacks the ids of a small number m of the most recent rumors y learned about, x can check to see if it is missing something) This combination of push rumoring and pull anti-entropy helps to reliably spread new information everywhere.
Retrieval: use vector space ranking, rank the relevance of document to the query by measuring the similarity of the query vector and document vector. TF IDF
