Estimation of overlay link quality from previously observed link qualities

To optimize overlay networks on the Internet, it is useful to estimate the qualities of each overlay link. Existing tools for measuring link qualities send probe packets from one end of the network to the other, but this requires the execution of measurement software at both end nodes. Besides, these tools need to send many probe packets to measure quality accurately, and this can disrupt other communication. In this paper, we propose a technique based on supervised learning for estimating link quality based on the quality observed for other similar links. Since our technique does not require sending probe packets, we can quickly estimate the qualities of many links without disrupting other communication or wasting processing time at many nodes. Through experiments on PlanetLab, the proposed method showed better performance on path latency estimation than the estimating results based on path length.