SDC: A Distributed Clustering Protocol

Network clustering is an important technique used in many large-scale distributed systems. Given good design and implementation, network clustering can significantly enhance the system's scalability and efficiency. However, it is very challenging to design a good clustering protocol for networks that scale fast and change continuously. In this paper, we propose a distributed network clustering protocol SDC targeting large-scale decentralized systems. In SDC, clusters are dynamically formed and adjusted based on SCM, a practical clustering accuracy measure. Based on SCM, each node can join or leave a cluster such that the clustering accuracy of the whole network can be improved. One big advantage of SDC is it can recover accurate clusters from node dynamics with very low message overhead. Through extensive simulations, we conclude that SDC is able to discover good quality clusters very efficiently.