Stable and Practical AS Relationship Inference with ProbLink
summary
Presents a probabilistic AS relationship inference system that actually reduces real-world errors: ProbLink is an EM-style, unsupervised classifier built from triplet, absence-of-path and vantage-point connectivity features, seeded by a three-step code leaf heuristic. Compared to AS-Rank it reduces error rates by 1.6–6.1× on hard link classes, increases route-leak detection precision and recall by 4.1× and 3.4×, reveals 27% more complex relationships, and yields about 81% precision / 68% recall on leak detection.
AI-written summary · about
chapters (12)
- 0:00Start
- 0:50AS Relationships
- 1:34Motivation
- 2:17Case Study: Route Leak Detection
- 6:02Outline
- 6:48A Simple Algorithm: CoreToLeaf
- 10:44Feature Design
- 11:08Triplet Feature
- 12:56Connectivity Features
- 15:24Accuracy Comparison
- 15:45Stability Analysis
- 16:33Practical Applications
related talks
29mSynthesis of dependable and self-driving communication networks
29mCorrelation is not causation! Or is it?
17mVerifying maximum link loads in a changing world
41mRerouting Flows when Links Fail
31mExpectAll: A BDD Based Approach for Link Failure Resilience in Elastic Optical Networks
19mAnyPro: Preference-Preserving Anycast Optimization based on Strategic AS-Path Prepending
30mClusterlink: Enabling Fine-Grained and High-Speed Application-Level Connectivity...
18mAPKeep: Realtime Verification for Real Networks