A dynamic threshold method of clustering update messages into routing events in BGP measurements

To better understand BGP dynamics, a number of previous measurement studies have focused on the convergence time of the routing events in the global routing system. A time-based approach has been developed to cluster BGP updates stream of the same prefix and the same observation point into routing events with threshold value. In practical networks, the same static threshold value, i.e. 4 minutes, can not group updates into events correctly. In this paper, we present an algorithm of dynamic threshold value for per destination network based on the probability method. Our experimental results show that the threshold value is sensitive to prefixes. A threshold value suitable for one prefix is not likely appropriate for another. Therefore, we think the dynamic values are more suitable for prefixes than the static value. With our method, the routing events can be clustered precisely, so that can improve the accuracy of the measurement studies.