Performance of search strategies in two-tier peer-to-peer networks

In this paper we analyze the performance of generalized flooding search strategies in two-tier peer-to-peer networks by means of random graphs. In particular, the connections between top and bottom-level peers are modeled as a bipartite random graph while the overlay network used by top-level peers to forward queries is modeled as a generalized random graph. To assign realistic values to the input model parameters we developed a distributed crawler of the Gnutella network to obtain snapshots of the application topology at both levels; to validate the model predictions we conducted simulation experiments on these snapshots. We exploited the model to show how generalizations of the classical flooding technique can help to design search strategies that meet pre-defined requirements on the hit probability and the average number of query messages exchanged during a search operation.