Computer Science editorial
Open AccessOA2026
Fake It No More: Evaluating L4S with SCReAM on Video Traffic
An IMRAD digest of L4S performance on SCReAM-controlled video over emulated mobile networks
Nawel Alioua; Ryan Zanone; Cheng Xi; Elizabeth Beldingยท 2026ยท DOI 10.48550/arXiv.2607.23767
The core problem
Interactive multimedia applications demand low latency, low loss, and scalable throughput. The L4S architecture addresses these needs by combining scalable congestion control with a dual-queue AQM. However, most evaluations focus on network-level metrics rather than end-user Quality of Experience (QoE). This paper asks: how does L4S affect SCReAM-controlled video traffic in terms of both network performance and QoE? The authors use an open-source DualPI2 implementation over the Mahimahi emulator, augmenting the SCReAM BW tool with a video codec to generate realistic video traffic. They evaluate a mobile network trace under random packet loss and different motion-complexity levels. The central finding is that L4S can reduce queue delay substantially at the cost of sender throughput, and that QoE gains are more pronounced under loss. The work underscores the need to assess QoE alongside network metrics when judging L4S for interactive video.
Innovation
In the baseline scenario (no packet loss), L4S reduces the median per-run 95th percentile queue delay by 35% compared to Classic. However, this comes at the cost of a 42% drop in sender throughput. Under 1% packet loss, L4S yields more pronounced QoE gains compared to the lossless scenario, despite narrower network-level benefits. Across video content complexities, L4S maintains more stable QoE than Classic. These results are summarized in the following table (values are approximate based on the abstract):
| Scenario | Metric | L4S vs Classic |
|----------|--------|----------------|
| Baseline (no loss) | Median 95th percentile queue delay | -35% |
| Baseline (no loss) | Sender throughput | -42% |
| 1% loss | QoE gains | More pronounced than lossless |
| 1% loss | Network-level benefits | Narrower |
| Varying motion complexity | QoE stability | L4S more stable |
The queue delay reduction can be expressed as:
And the throughput drop:
Interactive multimedia applications demand low latency, low loss, and scalable throughput. The L4S architecture addresses these needs by combining scalable congestion control with a dual-queue AQM. However, most evaluations focus on network-level metrics rather than end-user Quality of Experience (QoE). This paper asks: how does L4S affect SCReAM-controlled video traffic in terms of both network performance and QoE? The authors use an open-source DualPI2 implementation over the Mahimahi emulator, augmenting the SCReAM BW tool with a video codec to generate realistic video traffic. They evaluate a mobile network trace under random packet loss and different motion-complexity levels. The central finding is that L4S can reduce queue delay substantially at the cost of sender throughput, and that QoE gains are more pronounced under loss. The work underscores the need to assess QoE alongside network metrics when judging L4S for interactive video.
The authors implement L4S using an open-source DualPI2 AQM over the Mahimahi network emulator. They extend the SCReAM BW tool with a video codec, enabling the generation of video traffic in addition to its original synthetic RTP mode. The evaluation uses a mobile network trace, with scenarios including a baseline (no loss), 1% random packet loss, and different motion-complexity levels. Network-level metrics include queue delay (median per-run 95th percentile) and sender throughput. QoE metrics are derived from the video traffic, likely including stall duration, quality switches, and mean opinion score proxies. The experimental design allows comparison between L4S and Classic (likely a single-queue, loss-based or delay-based baseline) under identical conditions. The use of an emulator provides reproducibility while capturing realistic network dynamics. The augmentation of SCReAM with a video codec is a key methodological contribution, bridging synthetic and real video traffic.
Why it matters
The results reveal a fundamental trade-off: L4S achieves lower queue delay but reduces sender throughput. This may be acceptable for interactive video if QoE remains high, but the 42% throughput drop could harm applications requiring high bitrates. The finding that QoE gains are more pronounced under packet loss suggests L4S is particularly beneficial in lossy conditions, where Classic suffers from retransmissions and delay spikes. The stability across motion complexities indicates L4S handles varying video content better, possibly due to its scalable congestion control. However, the narrower network-level benefits under loss imply that L4S's advantage is not purely in delay reduction but in overall QoE. The authors conclude that evaluating QoE alongside network metrics is essential when assessing L4S for end-user applications. Future work could explore other video codecs, real-world traces, and additional QoE metrics. The study also highlights the importance of open-source implementations for reproducible research.
Who should read this
CS practitioners and researchers
Opening member contentโฆ