Effective Diagnosis of Routing Disruptions from End Systems

Internet routing events are known to introduce severe disruption to applications. So far effective diagnosis of routing events has relied on proprietary ISP data feeds, resulting in limited ISP-centric views not easily accessible by customers or other ISPs. In this work, we propose a novel approach to diagnosing significant routing events associated with any large networks from the perspective of end systems. Our approach is based on scalable, collaborative probing launched from end systems and does not require proprietary data from ISPs. Using a greedy scheme for event correlation and cause inference, we can diagnose both interdomain and intradomain routing events. Unlike existing methods based on passive route monitoring, our approach can also measure the impact of routing events on end-to-end network performance. We demonstrate the effectiveness of our approach by studying five large ISPs over four months. We validate its accuracy by comparing with the existing ISP-centric method and also with events reported on NANOG mailing lists. Our work is the first to scalably and accurately diagnose routing events associated with large networks entirely from end systems.