Interaction between Network Partitioning and Churn in a Self-Healing Structured Overlay Network

• Main Authors: Paul, Ruma R., Van Roy, Peter, Vlassov, Vladimir
• Format: Conference Proceeding
• Language: English
• Subjects: Complex systems; Computer simulation; Maintenance; Maintenance engineering; Mathematical models; Network Partition and Churn; Networks; Overlay networks; Partition Tolerance; Partitioning; Partitions; Peer-to-peer computing; Predictive models; Ring Overlay Merge; Routing; Strategy; Stress; Stresses; Structured Overlay Networks
• ISSN: 2690-5965; 1521-9097
• DOI: 10.1109/ICPADS.2015.37

We investigate the interaction between Network Partitioning and Churn (node turnover) in Structured Overlay Networks. This work is relevant both to systems with peaks of high stress (e.g., partitions, churn) or continuous high stress. It prepares the way for new application venues in mobile and ad hoc networks, which have high node mobility and intermittent connectivity, and undergo frequent changes in network topology. We evaluate existing overlay maintenance strategies, namely Correction-on-Change, Correction-on-Use, Periodic Stabilization, and Ring Merge. We define the reversibility property of a system as its ability to repair itself to provide its original functionality when the external stress is withdrawn. We propose a new strategy, Knowledge Base, to improve conditions for reversibility in the case of combined network partitioning and churn. By means of simulations, we demonstrate reversibility for overlay networks with high levels of partition and churn and we make general conclusions about the ability of the maintenance strategies to achieve reversibility. We propose a model, namely Stranger Model, to generalize the impact of simultaneous network partitioning and churn. We show that this interaction causes partitions to eventually become strangers to each other, which makes full reversibility impossible when this happens. Using this model, we can predict when irreversibility arrives, which we verify via simulation. However, high levels of one only, network partitioning or churn, do not hinder reversibility. In future work we will extend these results to real systems and experiment with applications that take advantage of reversibility.