[Paper Review] TCP Prague Fall-back on Detection of a Classic ECN AQM

Bob Briscoe, Asad Sajjad Ahmed|arXiv (Cornell University)|Nov 2, 2019

Network Traffic and Congestion Control22 references4 citations

TL;DR

This paper proposes a v2 passive detection algorithm for TCP Prague to detect classic ECN AQMs and fall back to Reno-friendly behavior, using adaptive RTT smoothing and delay variation metrics. The algorithm achieves near-perfect detection accuracy, reducing throughput imbalance between L4S and classic flows to a ratio of 0.5–2, satisfying IETF's Prague L4S coexistence requirement with minimal false positives.

ABSTRACT

The IETF's Prague L4S Requirements [DSBE20] expect an L4S congestion control to somehow detect if there is a classic ECN [RFB01] AQM at the bottleneck and fall back to Reno-friendly behaviour (as it would on a loss). This paper addresses that requirement in depth. A solution has been implemented in Linux and extensively tested, which distinguishes L4S from Classic AQMs primarily by their delay variation. This paper describes version 2 of the design of that solution, giving extensive rationale and pseudocode. It briefly summarizes a comprehensive testbed evaluation of the solution, referring to the full details online. The v2 algorithm very rarely falsely detects a Classic AQM as L4S. It also rarely detects an L4S AQM as Classic in the majority of scenarios, but not at low link rates and large RTTs. This report is a work in progress. It suggests ideas for improving on the approach. It also outlines new ideas that could solve the problem in complementary or alternative ways.

Motivation & Objective

• To address the IETF Prague L4S requirement for scalable congestion controls to detect classic ECN AQMs and fall back to Reno-friendly behavior.
• To ensure fair coexistence between L4S and classic ECN flows in shared-queue bottlenecks where classic ECN AQMs may exist.
• To design a passive detection mechanism that minimizes false positives while reliably identifying classic ECN AQMs across diverse network conditions.
• To evaluate and refine the detection algorithm for robustness across varying link rates, RTTs, and traffic mixes.
• To explore complementary active detection methods and future improvements, including out-of-band testing and marking spacing analysis.

Proposed method

• Uses passive detection based on RTT variation and ECN marking patterns to distinguish classic ECN AQMs from L4S AQMs.
• Employs adaptive exponential weighted moving average (EWMAs) for RTT smoothing with dynamic timescale adjustment based on RTT behavior.
• Applies a set of heuristics using inter-mark spacing and minimum RTT to detect classic ECN behavior.
• Implements a modular code structure with separate components for RTT smoothing, detection logic, and behavior transition.
• Introduces a fallback mechanism that triggers Reno-friendly congestion control when a classic ECN AQM is detected.
• Evaluates the algorithm in a comprehensive testbed with varied link rates, RTTs, and traffic mixes, including CoDel as the worst-case classic AQM.

Experimental results

Research questions

• RQ1 Can a passive detection algorithm reliably identify classic ECN AQMs based on delay variation and RTT behavior?
• RQ2 How does the v2 algorithm reduce false positives compared to v1, especially at low link rates and high RTTs?
• RQ3 To what extent does the algorithm ensure fair coexistence between L4S and classic ECN flows in shared-queue bottlenecks?
• RQ4 What are the performance trade-offs of using delay variation versus active probing for classic ECN detection?
• RQ5 Can out-of-band testing or marking spacing heuristics provide a more scalable and accurate alternative to passive delay-based detection?

Key findings

• The v2 algorithm reduces false positive detection of classic ECN AQMs to very low rates, especially in typical operating scenarios.
• The algorithm successfully triggers fallback to Reno-friendly behavior when a classic ECN AQM is present, ensuring throughput ratios between L4S and classic flows remain within 0.5 to 2.
• In the absence of the algorithm, throughput imbalances can reach up to 16:1, demonstrating the necessity of detection.
• False positives increase at link rates below 12 Mb/s and RTTs above 80 ms, indicating edge-case limitations.
• The algorithm performs robustly across diverse traffic mixes and is validated with CoDel as the worst-case classic AQM.
• Out-of-band testing and marking spacing heuristics are identified as promising alternatives for future validation with lower complexity.