Interdomain Routing

Some Foundational Problems in Interdomain Routing
  Autonomous Systems that route our packets to destinations have a tremendous difficulty in performing the packet routing between its peers. ASes do not cooperate well with each other in routing for many reasons : business motivation, traffic concern, or security. The state-of-the-art system still cannot solve the interdomain routing problem with clean hands.
  Interdomain routing problem reminds me of going on a hiking trail to an unfamiliar place. Because I do not know which path connects to where, I have to randomly choose a trail that looks popular and have a strong faith that it will lead me to the top. Even though BGP tries to fix the routing problem, the paper argues that BGP performs poorly because of its inability to validate the route, to dispute different AS policies, and to pick a converging path.
  Analyzing BGP's problems lead us to a better protocol that can ultimately serve our needs. The paper suggests that we can either add more information in the infrastructure like Griffin's work suggested or come up with a brand new protocol. More feasible solution like Gao and Rexford's approach would basically limit the business's policies and impair freedom in economic decision-making because ASes cannot mold into the exact role in real life. So it would not be a valid solution to this problem since the business would not want to use such approach for a long-term solution.
  Interdomain routing is a difficult area where scalability, subjectivity, and security problems all come across in designing a solution. However, the problem in BGP will multiply more and more as years go on. This paper helps to guide the solution to BGP's problems in near future.

Resolving Inter-Domain Policy Disputes
  Solving the problem of interdomain routing is one of many challenges the network field is facing today. Fortunately UCB has came up with an algorithm that solves the BGP's weaknesses and provides a new answer to the routing problem. In order to account for the policy-based routing, the algorithm accepts a globally enforced preference values to help stabilize the route chosen by ASes. The defining idea for solving the routing issue is this enforced preference values that each router advertise to each other. This value is then increased or decreased by the received router according to the algorithm that eventually gets rid of the oscillations.
  Another notable feature of the paper is the proof section on getting rid of the dispute wheel. Dispute wheel represents a presence of oscillation in the network, so the author gets rid of it by remembering the history of the preference values and modifying the preference value for each route according to the new advertisements received. It is in a sense similar to the path-vector algorithm used by BGP and thus make us wonder what's so different. The difference is that the values are all positive and that they eventually go down to 0 when the routes become stable.
  This noble idea is a distributed system that solves the routing problem. However I have a concern for this idea in that remembering the preference value for the paths can become a burden on that memory. If we introduce an abstraction layer to have nodes represent only ASes, we might be simplifying the network topology too much. Remembering these values is not scalable in my opinion and interdomain routing needs a good performance measure in scalability. Regardless, the paper is excellent in going one more step to analyze the nodes that are misbehaving with the system and how to identify them. This paper may have brought a new day to the interdomain routing.