L4Span: Spanning Congestion Signaling over NextG Networks for Interactive Applications

TL;DR

L4Span tackles the challenge of achieving end-to-end low latency for interactive applications over the cellular RAN by bridging L4S signaling with the 5G RAN. It introduces a pragmatic design that places signaling and marking logic inside the CU-UP, leverages live F1-U feedback, and uses per-UE queue-occupancy prediction to mark packets for both L4S and classic flows. The approach yields dramatic reductions in one-way delay (up to about 98% in RTT) while preserving near line-rate throughput across diverse transport protocols and channel conditions, and it maintains fairness across multiple UEs and DRB configurations. The findings suggest L4Span can be deployed with minimal RAN changes, is compliant with 3GPP/O-RAN, and improves interactive application performance in real-world 5G networks, potentially accelerating broader adoption of L4S in cellular networks.

Abstract

Design for low latency networking is essential for tomorrow's interactive applications, but it is essential to deploy incrementally and universally at the network's last mile. While wired broadband ISPs are rolling out the leading queue occupancy signaling mechanisms, the cellular Radio Access Network (RAN), another important last mile to many users, lags behind these efforts. This paper proposes a new RAN design, L4Span, that abstracts the complexities of RAN queueing in a simple interface, thus tying the queue state of the RAN to end-to-end low-latency signaling all the way back to the content server. At millisecond-level timescales, L4Span predicts the RAN's queuing occupancy and performs ECN marking for both low-latency and classic flows. L4Span is lightweight, requiring minimal RAN modifications, and remains 3GPP and O-RAN compliant for maximum ease of deployment. We implement a prototype on the srsRAN open-source software in C++. Our evaluation compares the performance of low-latency as well as classic flows with or without the deployment of L4Span in various wireless channel conditions. Results show that L4Span reduces the one-way delay of both low-latency and classic flows by up to 98 %, while simultaneously maintaining near line-rate throughput. The code is available at https://github.com/PrincetonUniversity/L4Span.

Figures (25)

• Figure 1: L4S status quo: Some backbone and broadband ISPs deploy L4S routers (left), but for mobile users, the 5G network is key to performance, yet lacks L4S functionality (right).
• Figure 2: Performance of L4S and CUBIC in different networks: L4S routers in wired networks achieve line rate and extremely low latency, but are much less effective in 5G networks that don't expose their queues. Both L4S and CUBIC's latency is reduced with L4Span in 5G, and both maintain line rate (green area). In the latter two figures, the bottleneck shifts from the RAN to wired middleboxes at 10s and shifts back at 20s.
• Figure 3: L4Span in the end-to-end data path from the content server to the client, where the red text and arrows inside the CU-UP and DU illustrate our system design, and the black elements inside the RLC entities mark where queues build up inside the 5G network. L4Span reuses the existing feedback in the RAN. L4S+L4S- denotes a router with/without L4S functionality; RU, MAC and PHY layers are not shown.
• Figure 4: L4Span's behavior on L4S and classic flows with wireless channel variations.
• Figure 5: L4Span's packet profile table based on the RAN feedback information. $T^{\{I, T,D\}}_{i}$ denotes the $i^{\mathrm{th}}$ packet's Ingress, Transmission, and Delivery timestamps. RU, MAC and PHY components are not shown.