Performance Analysis of Predictive Stabilization for Churn Handling in Structured Overlay Networks
Ramanpreet Kaur, Amrit Lal Sangal, Krishan Kumar · 2016
The structured overlay networks are self-organized, open networks with specific structural constraints to ensure resource or service discovery in a small number of probabilistically bounded networking hops. The main challenge to ensure the viability of the structured overlay networks for the deployment of highly scalable applications is to maintain the structure of these overlay networks in the presence of dynamic participants. The state of the art structured overlay networks employs a costly periodic stabilization mechanism for the maintenance of overlay structure and per node routing tables in the dynamic environments. However, many applications report the predictable dynamic behavior of different overlay nodes driven by underlying user's online behavior and its dependency on many factors such as, geographical position, time of the day, day of the week etc. This paper utilizes these characteristics for the implementation of an intelligent predictive stabilization mechanism for the structure maintenance and routing table updation in the chord based overlay networks. In this paper, we focus on the implementation and performance comparison of the Hidden Markov Model based and Neuro-Fuzzy based predictive stabilization techniques. The OverSim based simulation results clearly present that our proposed predictive stabilization techniques outperforms the existing periodic stabilization techniques in terms of better lookup success ratio and less maintenance overhead. Experimental results also demonstrate the suitability of Neuro-Fuzzy model a compared to HMM based prediction model for the failure prediction of dynamic nodes.