Recovering from Internet Routing Failures using Declarative Policies and Argumentation Protocols

Many Internet failures are caused by misconfigurations of the BGP routers that manage routing of traffic between domains. The problems are usually due to a combination of human errors and the lack of a high-level language for specifying routing policies that can be used to generate router configurations. We describe an implemented approach that uses a declarative language for specifying network-wide routing policies to automatically configure routers and show how it can also be used by software agents to diagnose and correct some networking problems. The language is grounded in an ontology defined in OWL[1] and polices expressed in it are automatically compiled into low-level router configurations. A distributed collection of software agents use the high-level policies and a custom argumentation protocol to share and reason over information about routing failures, diagnose probable causes, and correct them by reconfiguring routers and/or recommending actions to human operators. We have evaluated the framework in both a simulator and on a small physical network. Our results show that the framework performs well in identifying failure causes and automatically correcting them by reconfiguring routers when permitted by the policies.