An Optimized Distributed Clustering Algorithm in Advanced 3-layer Peer-to-Peer Network

In this paper, an advanced 3-layer Peer-to-Peer architecture and algorithm is introduced. The core of it is activated-peer of middle layer that is oriented on special P2P applications, such as large-scale network clustering and cooperation work. Our architecture is based on a multi-layer network of peer neighborhoods. Many activated-peers which act as management center of its lower neighborhoods are grouped to form upper level cluster. The level of activated-peer is dynamic which depends on neighborhood threshold set by root peer. Using this model, the clustering problem can be partitioned in an iterative way like solving each part individually in bottom peer which gets primitive data, then combine clustering up to upper activated-peer and repeat this method to the root peer of network. The matching algorithm is applied to a distributed document clustering problem of our architecture. It acts satisfied efficiency with comparable clustering quality to the centralized approach in the experiments.