Distributed Monitoring in Large Scale Overlay Networks (OverlayMeter)

点击翻译按钮获取中文摘要

Aiming at the quality of peer-to-peer overlay networks, research focused on optimized mechanisms for selected usage scenarios. The transfer of optimized solution from one specific use case to another is hardly or impossible to implement, as the parameter setting and strategy selection of the overlay is typically fixed for the specific use case. Either the overlays are blind for the dynamics of the environmental parameters, such as the node count or churn dynamics, and thus cannot adapt to a given situation. Or the various mechanisms that interact in the overlay themselves independently and redundantly try to gather relevant information for optimized decisions. In order to enable overlay networks to serve for a wide range of application areas with optimized quality, we aim in this project at providing of an information basis on the overlay which can be used for optimized overlay parameter settings and strategy selections. We consider in this information basis a set of statistics on the performance of the overlay, information on the capacities of the available nodes in the network as well as usage statistics on the documents in the overlay. The creation of future quality-centric overlays will be eased through the wide set of information available. New overlay mechanisms will be able to assume the existence of these information, interpret them and use them to optimally select parameter settings and strategies.In this research project we address the challenges in the creation of such a broad information basis on performance statistics in the overlay using only decentralized aggregation mechanisms. We investigate, design and evaluate efficient, precise and secure measurement mechanisms for overlay networks for the case that all nodes or only a subset of nodes participate in the measurement. The researched mechanisms should be applicable both on distributed hash tables as well as in general weakly connected overlay graphs which might also contain malicious nodes. Besides the focus on the creation of precise network-wide statistics on the performance of the overlay, we also focus on the research of measurement mechanisms to create a global view on the capacities of the peers in order to enable capacity-based peer search as well as on the usage statistics and rankings on the documents in the network. The strong dynamics of the peer capacities as well as the large number of the documents to be considered define new challenges for the measurement of these information.