Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Is it a Real CD Mismatch in Interdomain Routing?

Sun Letong, Shi Xingang, Han Fengyan, Yin Xia, Wang Zhiliang, Zhang Han

TL;DR

This work characterizes control-plane and data-plane CD mismatch in inter-domain routing using a large-scale, multi-year study with 128 vantage-point pairs. It introduces VISV to derive quasi-ground-truth IP-to-AS mappings from CD comparisons and presents LearnToCorrect, a learning-based method that markedly reduces mapping errors (about 0.5–0.8% intra-AS; ~70% of prior errors corrected) and lowers trace-level mismatch to around 11%. The authors find that real mismatches in the wild are typically under 6%, and demonstrate how CD-mismatch analysis can reveal routing security issues, such as hidden hijacks and bogus links. The study combines extensive data collection, multiple mapping strategies, and iterative evaluation to provide a robust framework for understanding CD mismatch and its implications for network operation and security; code and results are released to support reproducibility.

Abstract

In inter-domain routing, a packet is not always forwarded along the Autonomous System (AS) level path determined by the BGP routing protocol. This is often called control-plane and data-plane (CD) mismatch, which allows for flexible traffic control, but also leads to operation and security issues. We systematically analyze this phenomenon with path pairs collected from 128 pairs of vantage points over more than 5 years, and use multiple IP-to-AS mapping methods to compare CD paths. What is interesting is that, working at such a large scale in turn helps us design a novel method to fairly evaluate the accuracy of various existing mapping methods, and further develop a new mapping method, i.e., LearnToCorrect, that can correct more than 70\% mapping errors of the state-of-the-art one. Then we devise to identify real mismatches with LearnToCorrect, and estimate that the real-mismatch ratio in the wild is typically less than 6\%. At last, we use our proposed methods to detect routing security issues, which are previously difficult to accurately find out.

Is it a Real CD Mismatch in Interdomain Routing?

TL;DR

This work characterizes control-plane and data-plane CD mismatch in inter-domain routing using a large-scale, multi-year study with 128 vantage-point pairs. It introduces VISV to derive quasi-ground-truth IP-to-AS mappings from CD comparisons and presents LearnToCorrect, a learning-based method that markedly reduces mapping errors (about 0.5–0.8% intra-AS; ~70% of prior errors corrected) and lowers trace-level mismatch to around 11%. The authors find that real mismatches in the wild are typically under 6%, and demonstrate how CD-mismatch analysis can reveal routing security issues, such as hidden hijacks and bogus links. The study combines extensive data collection, multiple mapping strategies, and iterative evaluation to provide a robust framework for understanding CD mismatch and its implications for network operation and security; code and results are released to support reproducibility.

Abstract

In inter-domain routing, a packet is not always forwarded along the Autonomous System (AS) level path determined by the BGP routing protocol. This is often called control-plane and data-plane (CD) mismatch, which allows for flexible traffic control, but also leads to operation and security issues. We systematically analyze this phenomenon with path pairs collected from 128 pairs of vantage points over more than 5 years, and use multiple IP-to-AS mapping methods to compare CD paths. What is interesting is that, working at such a large scale in turn helps us design a novel method to fairly evaluate the accuracy of various existing mapping methods, and further develop a new mapping method, i.e., LearnToCorrect, that can correct more than 70\% mapping errors of the state-of-the-art one. Then we devise to identify real mismatches with LearnToCorrect, and estimate that the real-mismatch ratio in the wild is typically less than 6\%. At last, we use our proposed methods to detect routing security issues, which are previously difficult to accurately find out.
Paper Structure (25 sections, 12 figures, 9 tables)

This paper contains 25 sections, 12 figures, 9 tables.

Table of Contents

  1. Introduction
  2. Background and related work
  3. Compare control-plane and data-plane paths
  4. Data collection
  5. Data preprocessing
  6. IP-to-AS mapping methods
  7. Path pair comparison result
  8. Evaluate Mapping Methods
  9. Achieve ground truth from traceroute paths and by manual contact
  10. Achieve semi-ground truth from CD comparison
  11. Evaluate mapping methods with VISV
  12. An advanced mapping method
  13. Identify mapping errors
  14. Correct mapping errors
  15. Evaluation
  16. ...and 10 more sections

Figures (12)

  • Figure 1: Break paths into segments in CD comparison.
  • Figure 2: CDF of trace-level mismatch ratio under different mapping algorithms.
  • Figure 3: Workflow of VISV.
  • Figure 4: CDF of mapping error ratios of different mapping algorithms evaluated by VISV.
  • Figure 5: Results of the temporal adaptability test.
  • ...and 7 more figures