Myth: An Accurate and Scalable Network Coordinate System under High Node Churn Rate

Abstract

Network coordinate (NC) system is an efficient mechanism to predict Internet distance (latency) with limited times of measurement. In this paper, we focus on the node churn problem in distributed NC systems. Our study on Vivaldi, a representative distributed NC system, shows that under high node churn rate the prediction accuracy of the system will be seriously impaired, which renders the system unusable in many potential application scenarios. As remedy, we propose the idea and implementation of Myth, an accurate and scalable NC system even under high node churn rate. Myth introduces the merit of Landmark-based NC system to shorten convergence time in Simulation-based NC system with negligible extra overhead. We evaluate the performance of Myth system with two typical real-world data sets measured in Internet. The experimental results show that Myth outperforms Vivaldi in relative error, relative rank loss and closest neighbor absolute error under high node churn rate, without compromising the performance under stable environment.