Mr.Tree: Multiple Realities in Tree-based Monitoring Overlays for Peer-to-Peer Networks

摘要

Peer-to-peer systems are difficult to manage due to their decentralized nature and the lack of a central controlling instance. Thus, establishing a quality of service is a challenging task. In the last decade, decentralized monitoring approaches have been proposed, which are either precise or fault-tolerant. Tree-based approaches, which belong to the former category, are prone to churn as the structure needs to reorganize itself on node failures. In this paper, we propose the monitoring approach Mr.Tree, a protocol that enhances the robustness of existing tree-based approaches by introducing multiple realities, through building up multiple trees, and smartly replicating and maintaining aggregated data. Three modes are proposed to fit individual scenarios and needs. The evaluation shows a significant decrease of the error in completeness of monitoring data while keeping the costs relatively low. The error of the extended protocol under heavy churn is in the magnitude of the original protocol under low churn conditions, which is 8%, whereby the error of the original protocol under heavy churn raises to a value above 22%. The superior precision cancels out the increased costs already in scenarios with moderate churn.