An architecture for inter-domain network troubleshooting.

Abstract

Traditionally, network management mechanisms have concentrated on troubleshooting problems within a single administrative domain, where all devices are controlled by the same organization. As the Internet has evolved, however, more and more problems have effects reaching far beyond a single domain, and either troubleshooting has been manually coordinated, or such problems have been left unresolved or even undetected. As a result, problems abound in the current Internet and often take a significant amount of time to diagnose and repair. In this thesis, we build upon existing intra-domain network management techniques by introducing an architecture which allows automatic coordination of problem troubleshooting between domains. Many of the assumptions and constraints in network management are very different in the inter-domain context from those in the intra-domain context. Hence, this work addresses a new problem. We present a troubleshooting methodology that organizes troubleshooting activities into a unified theory. We have used this methodology to generate troubleshooting procedures and find deficiencies in existing mechanisms. We also provide a scalable protocol that operates at an inter-domain level and describe preliminary deployment experience with a prototype. This protocol, which we call the Globally-Distributed Troubleshooting (GDT) protocol, uses our methodology to troubleshoot network faults and congestion problems in the Internet. Finally, to support the GDT protocol, we also addressed the problems of automatic hierarchy construction and or efficiently selecting a server from a cluster. We provide solutions which are generally applicable to network protocols, and have already been used in other contexts such as multicast routing and World-Wide Web caching.