CISS

Abstract

Recently, Distributed hash tables (DHTs) have been widely adopted in many P2P systems as a scalable and efficient infrastructure. Since most of the recent DHT-based P2P systems perform key encoding using consistent hash functions, even semantically related objects are stored at randomly selected peer nodes. This de-clustering of semantically related objects results in significant inefficiency in processing multi-dimensional range queries and data updates as are common in emerging DHT-based applications such as massively multiplayer online games (MMOGs) and P2P auctions because of the large number of DHT lookups required for such processing. This paper proposes an efficient object clustering framework, CISS, for DHT-based P2P applications. Using DHT as a basic lookup layer, CISS applies a locality preserving function (LPF) for object clustering instead of a consistent hash function. Moreover, to maximize the benefit of object clustering, it proposes efficient routing protocols for multi-dimensional range queries and data updates. CISS also suggests efficient load balancing algorithms as the skewed distribution of queries and objects can result in significant load imbalance. The performance benefits of CISS are demonstrated by conducting intensive simulations. The results show that CISS has good load balancing performance and significantly reduces the number of DHT lookups compared to existing DHT-based P2P systems which use a hash function.