Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

IRR-Based AS Type of Relationship Inference

Amit Zulan, Omer Miron, Tal Shapira, Yuval Shavitt

TL;DR

The paper presents an IRR-based framework for inferring AS Type of Relationship (ToR) by mining AUT-NUM and AS-SET data across 23 IRR registries. It deploys three heuristics—Import-Export, Remarks, and Sets—along with reliability filtering and siblings inference to produce a large, high-confidence ToR dataset. The approach achieves high accuracy against manual ground truth (up to ≈95%) and uncovers many ToRs and siblings not present in prior datasets, increasing coverage while maintaining reliability. This work demonstrates IRR mining as a scalable source for ToR knowledge and provides a practical pathway for integrating IRR-derived ToRs into routing and security analyses.

Abstract

The Internet comprises tens of thousands of autonomous systems (ASes) whose commercial relationships are not publicly announced. The classification of the Type of Relationship (ToR) between ASes has been extensively studied over the past two decades due to its relevance in network routing management and security. This paper presents a new approach to ToR classification, leveraging publicly available BGP data from the Internet Routing Registry (IRR). We show how the IRR can be mined and the results refined to achieve a large and accurate ToR database. Using a ground truth database with hundreds of entries we show that we indeed manage to obtain high accuracy. About two-thirds of our ToRs are new, namely, they were not obtained by previous works, which means that we enrich our ToR knowledge with links that are otherwise missed.

IRR-Based AS Type of Relationship Inference

TL;DR

The paper presents an IRR-based framework for inferring AS Type of Relationship (ToR) by mining AUT-NUM and AS-SET data across 23 IRR registries. It deploys three heuristics—Import-Export, Remarks, and Sets—along with reliability filtering and siblings inference to produce a large, high-confidence ToR dataset. The approach achieves high accuracy against manual ground truth (up to ≈95%) and uncovers many ToRs and siblings not present in prior datasets, increasing coverage while maintaining reliability. This work demonstrates IRR mining as a scalable source for ToR knowledge and provides a practical pathway for integrating IRR-derived ToRs into routing and security analyses.

Abstract

The Internet comprises tens of thousands of autonomous systems (ASes) whose commercial relationships are not publicly announced. The classification of the Type of Relationship (ToR) between ASes has been extensively studied over the past two decades due to its relevance in network routing management and security. This paper presents a new approach to ToR classification, leveraging publicly available BGP data from the Internet Routing Registry (IRR). We show how the IRR can be mined and the results refined to achieve a large and accurate ToR database. Using a ground truth database with hundreds of entries we show that we indeed manage to obtain high accuracy. About two-thirds of our ToRs are new, namely, they were not obtained by previous works, which means that we enrich our ToR knowledge with links that are otherwise missed.

Paper Structure

This paper contains 25 sections, 6 equations, 7 figures, 11 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. IRR Structure
  5. Data Extraction and Pre-processing
  6. Heuristics
  7. Import-Export (I/E) Heuristic
  8. Remarks Heuristic
  9. Sets Heuristic
  10. Filtering Unreliable Data
  11. Siblings Inference
  12. Results
  13. Filtering Results
  14. Filtering Consistency
  15. Heuristics Union
  16. ...and 10 more sections

Figures (7)

  • Figure 1: An AUT-NUM object example.
  • Figure 2: Import-Export Heuristic Parameters Exploration
  • Figure 3: Remarks Heuristic Parameters Exploration
  • Figure 4: Sets Heuristic Parameters Exploration
  • Figure 5: CDFs of the number of ASes with the same declaration for each field used for siblings inference.
  • ...and 2 more figures