More Works from Our Lab

Promises and Potential of BBRv3

This paper focuses on characterizing the promises and potential of the third, and most recent, revision of BBR—introduced to the public in July 2023.

PAM 2024
Small Packets, Big Difference: L4S Congestion Control at Low Bandwidth

We propose and evaluate a pacing-rate-based segment sizing scheme for TCP Prague that mitigates RTT sensitivity and enhances performance, fairness, and robustness on low-rate bottlenecks.

ANRW 2026
Network Traffic Measurement and Analysis Conference (TMA) 2026

The State of ECN Adoption
in the Internet

Danesh Zeynali 1,2, Denis Eminov3, Taha Albakour 1, Balakrishnan Chandrasekaran 3, Anja Feldmann 1,2
1 Max Planck Institute for Informatics, Germany
2 Saarland University, Germany
3 Vrije Universiteit Amsterdam, The Netherlands
Paper Code

Abstract

Explicit congestion notification (ECN) is a signaling mechanism that enables a router along the network path between a sender and receiver to mark packets for signaling incipient congestion. More than a decade ago, Bauer et al. characterized the extent of ECN adoption on the Internet and a few years later Trammell et al. extended that analysis. Since then, the Internet has evolved substantially. Today, most operating systems support ECN, IPv6 has been widely deployed, and new technologies that build on ECN, such as L4S, have emerged. These changes naturally motivate the need to revisit the study of ECN adoption. We, hence, revisit and extend the two key ECN characterization efforts. We find that the majority of Internet service infrastructures are ECN-aware, although a nontrivial fraction of hosts, routers, and middleboxes bleach ECN bits. Many hyper-giants, unfortunately, bleach ECN bits along the path, which has substantial implications for the use of technologies such as L4S. In this study, we also note that IPv6 endpoints and paths generally show stronger support for ECN than those using IPv4.

Methodology Endpoint Support Network Support Longitudinal
Methodology

Methodology. We use both active and passive measurements. Our active measurements focus on ECN support by networks and end hosts, while our passive measurements study longitudinal ECN adoption across MAWI and M-Lab NDT dataset over years. More details →

Endpoint support for ECN

IPv4 endpoints support for ECN. Most web servers and endpoints are ECN-aware, and network elements along the path rarely interfere with ECN negotiation. We observe higher connection success rates and stronger ECN support over IPv6 than IPv4. More details →

Path support for ECN

Path support for ECN. Most paths preserve ECN-related ToS bits correctly, with more than 90% of the paths showing no bleaching. When bleaching occurs,often concentrated within a small set of ASes, and typically involves resetting ToS bits to zero. More details →

Longitudinal ECN adoption

Longitudinal ECN adoption. From the perspective of a transit network, ECN-enabled TCP remained a small fraction of flows but increased steadily since 2017. More details →

Methodology Endpoint Support Network Support Longitudinal

BibTeX

@inproceedings{zeynali2026ecn,
title = {The State of ECN Adoption in the Internet},
author = {Zeynali, Danesh and Eminov, Denis and Albakour, Taha and Chandrasekaran, Balakrishnan and Feldmann, Anja},
booktitle = {Network Traffic Measurement and Analysis Conference (TMA)},
year = {2026},
url = {https://inet-ecn.mpi-inf.mpg.de/}
}