Routing Management in Network Operations

Routing management is an integral aspect of network management. Routing implements connectivity, the core functionality of networks. It is also one of the most complex network functionalities. The high complexity of routing results from the distributed calculation of network routes, the need to accommodate network policies when calculating routes, and the requirement for routing to be robust to network dynamics. Given this complexity, routing operations are challenging, time-consuming, and error-prone.;In this thesis, we focus on two major problems that operators face in routing management, namely the routing visibility and the routing stability. Routing visibility refers to the knowledge of the routes that are used in data traffic forwarding, and routing stability refers to the frequency with which these routes change. Routing operations differ significantly according to the type of routing (intradomain vs. interdomain routing) and the type of network (enterprise vs. backbone network). We investigate these two problems in the following three operational environments: intradomain routing for enterprise networks, intradomain routing for backbone networks, and interdomain routing for backbone networks. We propose operationally viable solutions for routing visibility and routing stability in each network environment.;In enterprise intradomain routing management, we address the data path visibility problem caused by the numerous interdependent routing and switching mechanisms. In backbone intradomain routing management, we address the problem of tunnel path visibility when automated traffic-engineering is deployed. In backbone interdomain routing management, we address the problem of routing policy visibility caused by the numerous low-level routing policy specifications and the interdomain routing instability caused by the unstable interdomain routes advertised by the neighboring networks. The key idea behind the proposed solutions is the modeling of the network as a set of traffic pipes. Each network device realizes a segment of the pipe through which traffic flows with its routing or switching decision. The proposed solutions create models that abstract the network device functionality based on;the notion of traffic pipes. This abstraction model of the network functionality applies to all three operational environments.;Specifically, we address routing visibility through configuration analysis and lightweight implementations of the routing functionality. The configuration- and topology-based path inference system for enterprise networks enables detailed routing monitoring by abstracting the per-router routing functionality and the per-switch switching functionality. This system successfully infers the paths of 4000 packets in a set of data centers with complex routing design. The tunnel visibility system simulates the MPLS-TE functionality given the network's topology, the tunnels' configuration, and the tunnels' traffic demands. We verify that this system follows the real network operation by applying it on the backbone network of a major Internet service provider. The interdomain routing policy analysis system for backbone networks extracts the operators' high-level intent by abstracting the interdomain routing policies specified in router configuration. The application of this system on a production network generates six routing policy behaviors out of 128 routing policy specifications. Finally, we address the interdomain routing instability problem through automated reconfiguration based on routing state measurements. The proposed routing policy reconfiguration system increases the interdomain routing stability up to 50% by analyzing the history of interdomain route changes.;After comparing the intricacies of routing management in the three operational environments, we conclude that network configuration abstractions are essential for improved routing visibility and that configuration-based monitoring is essential for effective routing redesigns.