Internet Routing

Even today, given the widespread usage and critical importance of the Internet, its basic routing protocols such as the Border Gateway Protocol (BGP) are poorly understood. This is in part an artifact of the complex interactions that arise from a distributed system that is administered locally to achieve a global task: reachability. In another part it has its origin in the fact that inter-domain routing policies of autonomous systems (ASes) often undergo constant adjustments for reasons of traffic engineering and/or to address specific customer wishes, an error prone approach.

In my recent work I approached these issues in the following ways:

The problem of policy configurations by developing a system that allows us to manage the overall routing architecture rather than each individual router. With this we raise the abstraction level from individual BGP configuration statements to an AS-wide routing policy. Initial deployment of the system to manage the routing policy of Deutsche Telekom affirm the advantages in an operational setting.

The richness of policy expressions leads to complex interactions and dynamics that can be observed throughout the Internet. We developed a clustering technique for a BGP alarm system that helps operators to detect problematic routing conditions. This includes a methodology for identifying the AS that is responsible for an instability.

While we find that some of todays routing issues are stemming from protocol interactions, others are coming from router software/hardware problems that should have been detected in a test-lab before deployment in the operational network. In joint work with Agilent Labs we develop a BGP workload generator that can help in a wide variety of equipment testing. Its goal is to be able to instantiate complex tests without losing the test engineer in the intricacies of the test setup.