Shedding light on the use of AS relationships for path inference

Autonomous system (AS) business relationships and their inference have been widely studied by network researchers in the past. An important application of inferred AS relationships can be the prediction of AS paths between a source and destination AS within a model. However, besides knowing the topology and inferred AS relationships, AS path prediction within a model needs to be understood in order for us to know how we can derive border gateway protocol (BGP) policies from AS relationships. In this paper, we shed light onto the predictive capabilities of AS relationships by investigating whether they can be translated into BGP policies such that inferred AS paths are consistent with real AS paths, e.g., paths observed from BGP routing tables. Our findings indicate that enforcing constraints such as the well-known valley-free property and the widely assumed preference of customer routes always results in a very low consistency for AS path inference. In addition, this is true irrespective of whether customer, peer, or provider routes are preferred. Apparently, applying such constraints eliminates many “correct” paths that are observed in BGP routing tables and that are propagated in a simple shortest path model where AS relationships are ignored. According to our findings, deriving BGP routing policies for predicting with high accuracy AS paths in a model directly from AS relationships is still difficult.