A proximity-based hierarchical overlay framework for distributed hash tables

Abstract

Several structured peer-to-peer networks have been created to solve the scalability problem of previous peer-to-peer systems such as Gnutella and Napster. These peer-to-peer networks which support distributed hash table functionality construct a sort of structured overlay network, which can cause a topology mismatch between the overlay and the underlying physical network. To solve this mismatch problem, we propose a proximity-based hierarchical overlay framework for Distributed Hash Tables. The hierarchical approach for the overlay is based on the concept that the underlying global Internet is also a hierarchical architecture, that is, a network of networks. This hierarchical approach for the overlay puts forth two benefits: finding data in a physically near place with a high probability, and smaller lookup time. Our hierarchical overlay framework is different from other hierarchical architecture systems in a sense that it provides a concrete self-organizing grouping algorithm. This kind of self-organization makes our framework more suitable for pure peer-to-peer concepts. Basic design of our system consists of node join algorithm, data publication algorithm, and data retrieval algorithm. To verify how efficient the overlay is we suggest three metrics - data publication time, data lookup time and direct delay for data retrieval. To improve these three metrics is the goal of our system. All these three metrics depend not only on how near the nodes group together in a sub-network but also on how well balanced the size of the super-network and the sub-network is. Our additional improvement extension schemes enhance these three metrics and complete the basic algorithm which constructs a hierarchical structure without any central control.