Dynamic Semantic Data Replication for K-Random Search in Peer-to-Peer Networks

Abstract

We present a dynamic semantic data replication scheme called DSDR for classic k-random search in unstructured peer-to-peer (P2P) networks. During its k-random search each peer periodically updates its local view on the semantic overlay of the network based on observed queries (demand) and received information about provided items (supply), in particular their semantics. Peers dynamically form potentially overlapping groups for semantically equivalent or similar items they are actually demanding. Besides, each peer predicts the number of needed item replicas in the future based on its local observations in the past. The decision of which item to best replicate to which member is made within each demander group based on the maximal expected utility, traffic costs, and plausibility of such replication. Our experimental evaluation evidences that k-random search with DSDR-based replication can significantly outperform its combination with a near-optimal but non-semantic replication strategy, as well as a peer expertise-based semantic P2P search without replication.