Endpoint Cluster Identification for End-to-End Distance Estimation

Distributed systems such as peer-to-peer networks and distributed servers can optimize their performance by adapting to the underlying network. End-to-end measurements are an important basis for such adaptivity. Although most applications measure similar properties of the network, the measurements are mostly done in application-specific ways. In this paper we propose a general peer-to-peer measurement service based on clusters of endpoints that show virtually identical QoS properties when observed from outside the cluster. We discuss the clustering concept as well as its use in the measurement service, and we present a measurement-based method for the remote identification of clusters. This method allows for detecting clusters that are not part of the peer-to-peer network. Our evaluation shows that the presented method is able to reliably detect clusters using measurements of round-trip time or of available bandwidth.