Abstract
Traditional network topology inference aims at reconstructing the routing trees rooted at each probing source from end-to-end measurements. However, due to emerging technologies such as network function virtualization, software defined networking, and segment routing, many modern networks are capable of supporting generalized forwarding that can create complex routing topologies different from routing trees. In this work, we take a first step towards closing this gap by proposing methods to infer the routing topology (referred to as 1-1-N topology) from a single source to multiple destinations, where routes may be required to traverse a given waypoint. We first thoroughly study the special case of 1-1-2 topologies, showing that even this seemingly simple case is highly nontrivial with 36 possibilities. We then demonstrate how the solution to the special case can be used as building blocks to infer 1-1-N topologies. The inferred topology is proved to be equivalent to the ground truth up to splitting/combining edges in the same category.