Advancements in the Inference of AS Relationships Xenofontas Dimitropoulos (Fontas) (CAIDA/GaTech) Dmitri Krioukov Bradley Huffaker k claffy George Riley
2 Outline Background & previous works. Background & previous works. Limitations of existing algorithms. Limitations of existing algorithms. New algorithm to infer customer to provider (c2p) peer to peer (p2p) and sibling to sibling (s2s) relationships. New algorithm to infer customer to provider (c2p) peer to peer (p2p) and sibling to sibling (s2s) relationships.
3 Motivation Evolution of the Internet. Evolution of the Internet. –What are the economics of the Internet and their role in shaping it? Connectivity doesn’t mean reachability. Connectivity doesn’t mean reachability. –AS relationships determine the valid paths along which packets can flow. Finding a provider or peer for ISPs. Finding a provider or peer for ISPs. –Details of peering relationships aren’t widely available. –Want to make a more informed decision based on actual relationships and connectivity instead of “brand name”. Determining AS rank. Determining AS rank. –What is the relative importance of the Internet players? –Useful for determining order for security notifications.
4 Valley-free model AS paths have the following hierarchical structure: AS paths have the following hierarchical structure: 1.zero or more customer-provider (or sibling) links. 2.zero or one peer-peer link. 3.zero or more provider-customer (or sibling) links. GAO: “On Inferring Autonomous Systems relationships in the Internet” L. Gao ACM IEEE Transactions on Networking Customer-provider link Customer-provider links Peer-peer link
5 Previous Works GAO GAO –Based on hierarchical property of AS paths. –Use AS degree to find top of the hill and assign transit relationships to the side-hill links. SARK SARK –Introduce the Type-of-Relationships (ToR) problem: Given an undirected graph G derived from a set of BGP paths P, assign the edge type (customer-provider or peering) to every edge in G such that the total number of valid paths in P is maximized. –Provide a heuristic solution. SARK: “Characterizing the Internet hierarchy from multiple vantage points” L. Subramanian et. al INFOCOM 2002.
6 Previous Works DPP and EHS: DPP and EHS: –No peering can be inferred in ToR. –ToR is NP-complete (by reducing to MAX2SAT). –Mathematically rigorous approximations to ToR. DPP: "Computing the types of the relationships between Autonomous Systems", G. Di Battista et al. IEEE INFOCOM EHS: “Classifying customer-provider relationships in the Internet”, T. Erlebach et al. IASTED CCN 2002.
7 ToR limitations ToR does not provide a deterministic solution, because some links can be directed either way and produce the same number of valid paths. ToR does not provide a deterministic solution, because some links can be directed either way and produce the same number of valid paths. ToR solutions may treat anomalous paths as valid, which results in incorrect AS relationships. ToR solutions may treat anomalous paths as valid, which results in incorrect AS relationships. –BGP misconfigurations and complex polices can violate the valley free model
8 Our customer to provider algorithm Generate Input: Generate Input: 1.Produce AS graph (with no relationships). 2.Assign weight to edges based on node degree. Formulate c2p inference as a multiobjective optimization problem: Formulate c2p inference as a multiobjective optimization problem: –O1: Maximize number of valid paths. –O2: Maximize the sum of weights over all edges. Intuition: Intuition: –Paths are now colored by their edges weights which allows to detect anomalous paths Valid with small weight Invalid with large weight W = 0 W > 0
9 Work in progress Inference of p2p links. Inference of p2p links. Inference of s2s links. Inference of s2s links. Validation of links. Validation of links. Create AS Rank based on AS relationships. Create AS Rank based on AS relationships.
10 Thanks! ? Inferring AS Relationships: Dead End or Lively Beginning? 4th Workshop on Efficient and Experimental Algorithms (WEA),