An Architecture for Inter-Domain Troubleshooting (extended version)

In this paper, we explore the constraints of a new problem: that of coordinating network troubleshooting among peer administrative domains or Internet Service Providers, and untrusted observers. Allowing untrusted observers permits any entity to report problems, whether it is a Network Operations Center (NOC), end-user, or application.

Our goals here are to define the inter-domain coordination problem clearly, and to develop an architecture which allows observers to report problems and receive timely feedback, regardless of their own locations and identities. By automating this process, we also relieve human bottlenecks at help desks and NOCs whenever possible.

We begin by presenting a troubleshooting methodology for coordinating problem diagnosis. We then describe GDT, a distributed protocol which realizes this methodology. We show through simulation that GDT performs well as the number of observers and problems grows, and continues to function robustly amidst heavy packet loss.