BGP route prediction within ISPs

Abstract

Internet Service Providers (ISPs) often collect routing data to troubleshoot, analyze and predict the behavior of their network. However, data collected from multiple interacting routing protocols is often incomplete and difficult to manually analyze. In this paper we present a systematic approach to combine the pieces of measured routing data to obtain a more complete picture of a network's routing state. Our technique is efficient, has no assumptions about router configuration and is accurate. We present a case-study of a large Tier-2 ISP, finding that for those routers with adequate measurement infrastructure, we consistently find the egress location for 99.9999% of (router,prefix) pairs. Further, for the 85% of routers without measurement infrastructure we predict their decisions. This technique has been successfully applied in a 'what-if' scenario and has future applications in the real-time analysis of routing decisions.©2010 Elsevier B.V. All rights reserved.