POND: The Power of Zone Overlapping in DHT Networks

Abstract

Distributed hash table (DHT) based algorithms have been proposed to solve the routing issues in large-scale peer-to-peer (P2P) applications. DHT systems create an elegant peer-to-peer service model and work perfect in a homogeneous environment. Workload are evenly distributed on every participating peer and system is supposed to achieve the optimal performance. However, in the real world, the complicated scenarios such as peerspsila capability heterogeneity, diverse file popularities, and mismatching problem between physical network structure and logical routing data structure, jeopardize the efficiency of DHT infrastructure and may result in serious load imbalance problem. In this paper, we propose -- POND, a novel and efficient DHT algorithm to relieve the above problem. In POND, peers are divided into two types: main peers (MP) and shadow peers(SP). As other DHT systems, POND splits the name space into several disjointed zones. However, unlike other systems which only have one peer in charge of each zone, in POND, we use zone overlapping technique. In each zone, a MP and several SPs are used for services. Any of them is capable to handle service requests fallen into this zone. When a new request is coming,the peer with the minimal load has the high possibility to be chosen to provide the service. Thus, the risk of overload problem is reduced. With the introduction of shadow peers, POND also provides better failure tolerance and relieves the hot spot problem. Moreover, POND can improve the routing performance by reducing the average number of routing hops per request. We conduct extensive simulation experiments and the results show POND is superior to other DHT algorithms.