SkyEye: A tree-based peer-to-peer monitoring approach

Peer-to-Peer overlays, used by applications like distributed online social networks, need the ability to adjust their system parameters and thus the system performance to provide a certain level of quality, i.e. to ensure maximum response times. For the control of p2p network, a reliable information basis about the system's state is essential, i.e. what the current average response time is. A precise and rapidly adapting monitoring mechanism can provide the desired knowledge. In this paper we propose SkyEye, a tree-based monitoring approach which operates on top of an existing structured p2p overlay. It provides through a lightweight, error-prone protocol a continuous monitoring of all peers in the network. It gathers and disseminates sophisticated statistics on a wide range of metrics from all peers, to all peers. Evaluation shows a superior precision and adapting rate compared to an implementation of the gossip-based push-sum algorithm in scenarios with churn and without churn.