The resilience of the Internet is largely based on the diversity of connectivity between its networks. Such diversity means that when one route between you and the network you’re trying to access is not working, you can take another route to your destination.
Researchers and network operators constantly monitor and measure these routes to understand their performance, troubleshoot issues, find optimal customer routes, and map the Internet. One way they do this is by using a traceroute — a command that sends small packets to a target destination and measures the latency of the multiple routes they take.
While this tried and tested method provides insight into the performance of the routes between your network and the network you’re trying to connect to, it doesn’t consider the route/s the packets take back to your network.
That’s why researchers and network operators also need to use a reverse traceroute tool to measure routes to their own hosts from arbitrary networks without access to those remote networks to run the command.
Reverse Traceroute + Network Diagnostic Tool = Improved Visibility
revtr 2.0 is a second-generation reverse traceroute tool that combines novel measurement approaches and studies with a large-scale deployment to improve throughput, accuracy, and coverage, enabling the first exploration of reverse paths at Internet scale.
Recently, my colleagues and I from LAAS-CNRS, Columbia University, Universidade Federal de Minas Gerais, and Northeastern University collaborated with M-Lab to include revtr 2.0 in its Network Diagnostic Tool (NDT).
NDT allows a user to measure its bandwidth and latency to an M-Lab server. As part of each NDT test, M-Lab issues a forward traceroute from the M-Lab server to the user to gather forward path information that they make publicly available to the community and also now provides reverse-path information via a revtr-sidecar, which triggers a reverse traceroute from the user back to the M-Lab server.
My colleagues and I have continuously run reverse traceroutes for a fraction of NDT tests (~500,000 reverse traceroutes a day) since 1 November 2023.
Based on our previous research, we’ve observed how revtr 2.0 can:
- Measure at least one reverse path from 39,544 Autonomous System (AS) destinations, which host 92.6% of Internet users. As a comparison, RIPE Atlas vantage points can measure from 4,344 AS destinations, representing 67.1% of the Internet user base.
- Return paths that are correct at AS level at least 98.3% of the time on our evaluation dataset.
- Measure ≈15M reverse and forward traceroutes per day during the collection of our large-scale campaign, or about 173 reverse traceroutes per second overall. revtr 2.0 sends 26% as many probes as revtr 1.0, corresponding to a 43x throughput (which is due to how we must monitor the spoofed packets) — see the paper for more details.
Improved Visibility = Improved Performance and Understanding of Internet
Another part of our research has been to show the practicality of revtr 2.0 for network operators and researchers.
For operators, we showed how revtr 2.0 could be integrated into traffic engineering systems to help identify the cause of a client experiencing poor performance. We announced our own prefixes on the Internet as if we were a network operator with multiple points of presence and measured the latency between destinations representing clients and our servers. We then ran reverse traceroutes from the clients experiencing poor latency to find the network on the path responsible and changed our announcement to shift them to a better path, improving their latency. In another experiment, we also showed how to use revtr 2.0 to perform load balancing — see our paper for more details.
For researchers, we performed the most extensive symmetry study of paths in the Internet, showing that 47% of the routes of the Internet were asymmetric at the AS level (not sharing the same reverse path).
If you’re interested in learning more about our study, read our paper, Internet Scale Reverse Traceroute, or join the system. All our data collected with M-Lab is publicly available on BigQuery. You can also email us at [email protected] to add your own source to the system.
Kevin Vermeulen, a full-time CNRS researcher at LAAS, is interested in designing and implementing measurement systems to build a better Internet.
The views expressed by the authors of this blog are their own and do not necessarily reflect the views of the Internet Society.