Yi Wang will present his preFPO on Wednesday December 3 at 1:30 PM in Room 401. The members of his committee are: Jennifer Rexford, advisor; Michael Freedman and Kobus van der Merwe (AT&T), readers; Larry Peterson and Ed Felten, nonreaders. Everyone is invited to attend his talk. His abstract follows below. ------------------------------------ Title: A Principled Approach to Managing Routing in Large ISP Networks Abstract: Today's Internet Service Providers (ISPs) face a number of challenges in introducing changes to their networks at different frequencies. Firstly and most frequently, ISPs need to perform daily planned maintenance to keep their networks running healthily. However, despite of the importance and frequency of the planned maintenance operation, there is no existing method that can handle such operations without causing packet losses and routing protocol re-convergence. Secondly, once in a while, ISPs have to adjust their routing policies to meet changing needs and make flexible trade-offs between different policy objectives (such as business relationships, performance, security, etc.). However, today's network architecture forces administrators to realize network- wide routing policies through router-level configurations and makes it very difficult (if not impossible) to make flexible trade-offs. Thirdly, ISPs also have the incentive to make a one-time change (transition) from the current restrictive route-selection model, in which each router selects a single best route for all neighbors, to a neighbor- specific route-selection model, in which each router can provide value-added customized route- selection services to different neighbors. However, it is currently unknown if such transition to a more flexible route-selection model may cause the Internet routing to oscillate. In this thesis, we take a principled approach to solving these "change-related" challenges and propose three abstractions that guide the design and implementation of our system solutions. First, we propose the abstraction of a "neighbor-specific route selection problem" and a corresponding "Neighbor-Specific BGP" (NS- BGP) model that capture the requirement of customized route selection for different neighbors. We prove that the Internet is guaranteed to remain stable even if individual ISPs make transition to this more flexible route-selection model incrementally. Second, we propose the abstraction of "policy configuration as a decision problem of reconciling multiple objectives". Guided by this abstraction and the Analytic Hierarchy Process, a decision-theoretic technique for balancing conflicting objectives, we designed and implemented a prototype of an extensible routing control platform (Morpheus) that enables an ISP to select routes for different neighbors individually and make flexible trade-offs among policy objectives through a simple and intuitive configuration interface. Finally, we propose the abstraction of the separation between "physical" and "logical" configurations of routers, which leads us to the design and prototype implementation of "virtual router migration" (VROOM), a new, generic technique to simplify/enable a broad range of tasks that require the network to adapt in certain ways, from planned maintenance to reducing power consumption. Collectively, the contributions of the thesis provide simple and effective solutions for an ISP to autonomously introduce changes to its network at various frequencies without sacrificing global routing stability.