Measuring routing dynamics induced by the AS path prepending method

Abstract

Thousands of autonomous systems (ASes) connect to each other to form the Internet today. They exchange reachability information via the only inter-domain routing protocol, Border Gateway Protocol (BGP). BGP provides attributes for individual ASes to express their routing preferences which are often a result of the routing policies. Moreover, an increasing number of ASes connect to more than one AS in order to facilitate multi-homing. Thus, inbound traffic engineering has become a crucial task for network operators. Among a handful of inter-domain inbound traffic engineering methods, AS path prepending is a widely practiced method which provides network resilience and does not increase routing table size. Unfortunately, network operators often perform prepending on a trial-and-error basis, which can lead to suboptimal results and a large amount of network churn. This dissertation studies the effects of the AS path prepending method based on Internet measurement. In particular, we have developed an active measurement methodology to study the prepending method. In this method, we actively inject prepended routes into the Internet routing system. We have implemented this method in two networks. One is on the RIPE Network Coordination Centre (NCC) Routing Information Service (RIS) which is one of five Regional Internet Registries (RIRs). The other is a Hong Kong local campus network. We have observed the resulting changes from almost 200 publicly-accessible sources of BGP information. Our results show that the measurement methodology is scalable and effective to study the effects of prepending announced by a stub AS which uses prepending to control its inbound traffic. Our analysis also shows that a small number of ASes is often responsible for a large amount of route changes induced by prepending. Furthermore, our methods are able to reveal hidden prepending policies and tie-breaking decisions made by ASes. These new observations and insights are useful for further predicting the effectiveness of the prepending method.