Towards Sub-millisecond Latency and Guaranteed Bit Rates in 5G User Plane CCS: Networks Programmable networks CCS: Networks Network performance modeling
Leonardo Alberro, Noura Limam, Raouf Boutaba · 2025
Read Towards Sub-millisecond Latency and Guaranteed Bit Rates in 5G User Plane CCS: Networks Programmable networks CCS: Networks Network performance modeling by Leonardo Alberro, Noura Limam, Raouf Boutaba (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-00196.txt.
Introduction
Leonardo Alberro, Noura Limam, and Raouf Boutaba present a quality-of-service-aware data plane model for programmable transport networks serving the 5G user plane. Next-generation services demand per-flow bandwidth assurance, sub-millisecond-style latency goals, and traffic prioritization, while network functions migrate to edge and central clouds, making transport the critical enabler of end-to-end guarantees. Fixed-function infrastructure cannot flex across the diverse profiles standardized for 5G.
Their model supports predictable behavior and fine-grained service differentiation in programmable hardware. Telecommunications readers get a design for meeting strict service guarantees in cloudified networks. Predictability, not average speed, is what industrial and emergency services actually buy.
Catalog record
| CREATOR | Leonardo Alberro, Noura Limam, Raouf Boutaba |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2511-00196
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Towards Sub-millisecond Latency and Guaranteed Bit Rates in 5G User Plane CCS: Networks Programmable networks CCS: Networks Network performance modeling Leonardo Alberro, Noura Limam, Raouf Boutaba Abstract . Next-generation services demand stringent Quality of Service (QoS) guarantees, such as per-flow bandwidth assurance, ultra-low latency, and traffic prioritization, posing significant challenges to 5G and beyond networks. As 5G network functions increasingly migrate to edge and central clouds, the transport layer becomes a critical enabler of end-to-end QoS compliance. However, traditional fixed-function infrastructure lacks the flexibility to support the diverse and dynamic QoS profiles standardized by 3GPP. This paper presents a QoS-aware data plane model for programmable transport networks, designed to provide predictable behavior and fine-grained service differentiation. The model supports all 3GPP QoS resource types and integrates per-flow metering, classification, strict priority scheduling, and delay-aware queuing. Implemented on off-the-shelf programmable hardware using P4 and evaluated on an Intel Tofino switch, our approach ensures per-flow bandwidth guarantees, sub-millisecond delay for delay-critical traffic, and resilience under congestion. Experimental results demonstrate that the model achieves microsecond-level latencies and near-zero packet loss for mission-critical flows, validating its suitability for future QoS-sensitive applications in 5G and beyond. 1. Introduction Mobile networks are undergoing a paradigm shift driven by the need to accommodate diverse services with varying performance requirements, broadly categorized as enhanced Mobile Broadband (eMBB), massive Machine Type Communications (mMTC), and Ultra-Reliable Low Latency Communications (URLLC). From bandwidth-heavy real-time gaming and high-definition video streaming to massive IoT deployments and mission-critical control signaling, each service class demands specific latency, throughput, and reliability guarantees. Network slicing in 5G networks leverages virtualization to partition a shared physical infrastructure into multiple logical networks (slices), each tailored to a particular service class and Quality of Service (QoS) profile ( 3GPP 2024a ; Kazmi et al. 2019 ) . By decoupling resources and configurations end-to-end across the RAN, transport, and core segments, slicing enables eMBB, mMTC, and URLLC services to coexist without cross-interference. It promises guaranteed bit rates, ultra-low delays, or massive connection capacity as required. Recent research efforts have significantly advanced the understanding and implementation of slicing in both the core ( Ye et al. 2018 ; Li et al. 2019 ) and RAN ( Balasingam et al. 2025 ; Kokku et al. 2012 ; Chen et al. 2023 ; Foukas et al. 2017 ) segments, contributing toward an end-to-end (E2E) slicing architecture for 5G. As 5G network functions migrate toward cloud-native architectures, the transport network is increasingly critical in maintaining QoS guarantees. Traditional transport infrastructures lack the flexibility needed to support modern services’ dynamic and heterogeneous requirements. Therefore, a shift toward programmable data planes has emerged as a promising solution, enabling precise and scalable traffic management without compromising throughput or latency ( Bosshart et al. 2013 ; Xing et al. 2022 ; Chen et al. 2024 ; Yang et al. 2024 ; Xing et al. 2023 ) . Technologies such as the P4 programming language ( Bosshart et al. 2014 ) , combined with programmable hardware platforms like Tofino switches and smartNICs ( Corporation 2025 ; Netronome 2024 ) , provide the necessary building blocks for this evolution. In this context, a few studies have explored programmable QoS enforcement in the 5G transport (5GT), including mechanisms for per-slice bandwidth guarantees ( Chen et al. 2022 ) , delay-aware scheduling ( Harkous et al. 2021 ) , and QoS-oriented data planes ( Ricart-Sanchez et al. 2020 ) . However, these approaches often suffer from trade-offs, such as sacrificing bandwidth for delay guarantees, and lack support for dynamic prioritization, which is mandated by 3GPP ( 3GPP 2024b ) . Furthermore, they typically address slice-level requirements without accommodating the finer granularity of individual flow-level QoS. We argue that the current state-of-the-art lacks a comprehensive QoS-aware transport data plane that fulfills next-generation services’ diverse and demanding requirements. Specifically, a solution is needed that can simultaneously enforce guaranteed bit rates, deliver ultra-low latency, and dynamically prioritize traffic under network stress, all in alignment with 3GPP’s standardized flow-level QoS profiles. This paper introduces QoS-driven data plane forwarding behaviours to ensure predictable per-flow QoS in 5G transport networks. Our forwarding behaviours slice the 5G transport based on standardized QoS profiles, mapping flows to specific resource types, priority levels, and delay budgets. We validate our design on our in-lab high-speed 5G testbed equipped with a P4-programmable Intel Tofino switch ( Corporation 2025 ) . We implement our packet handling and forwarding behaviours in the switch using a combination of P4-based techniques, such as meters, traffic shaping, packet scheduling, and multiple hardware queues, demonstrating our solution’s ability to enforce per-flow bandwidth guarantees under congestion, sustain ultra-low delays for delay-sensitive traffic, and prioritize packets during network congestion. In summary, our contributions are: • A framework for QoS-aware 5GT slicing with programmable hardware, mapping 3GPP 5QI values to four distinct resource classes: Guaranteed Bit Rate (GBR), Delay-Critical Guaranteed Bit Rate (GBR*), Non-Guaranteed Bit Rate (Non-GBR), and Delay-Critical Non-Guaranteed Bit Rate (Non-GBR*). • A programmable data plane model with tailored packet handling and forwarding behaviours that provides per-flow bandwidth guarantees, ensures sub-millisecond latency for delay-critical traffic, and enforces packet prioritization during congestion. • A practical implementation and experimental evaluation using a real high-speed ASIC, supported by a custom 5G user-plane traffic generator with configurable QoS parameters. The rest of this paper is organized as follows. Section § 2 presents background, motivation, and related work. Section § 3 defines the foundations for a QoS-aware 5GT. Section § 4 describes our model design and analysis. Section § 5 details the implementation and evaluation results. Finally, Sections § 6 and § 7 discuss future works and conclude the paper. 2. Background and Related Work This section introduces fundamental concepts underlying QoS management in 5G networks. We begin by outlining the 3GPP-defined QoS model, which provides the framework for supporting heterogeneous service requirements. We then present an overview of data plane programmability with P4, emphasizing its role in enabling fine-grained QoS enforcement. Finally, we discuss prior work on network slicing and QoS-aware forwarding, focusing on the emerging use of programmable data planes to meet the stringent demands of next-generation services. 2.1. 5G Services and QoS Requirements To address the distinct needs of 5G use cases, such as eMBB’s high throughput and capacity, mMTC’s large number of low-data-rate connections, and URLLC extremely low latency and high reliability in support of mission-critical applications, 3GPP introduces the concept of QoS flows , which represent the finest level of QoS granularity in the 5G ecosystem ( 3GPP 2024b ) . Each QoS flow is associated with a 5G QoS Identifier (5QI) that dictates specific requirements from the network. 5QIs are mapped into a resource type, e.g., Guaranteed Bit Rate (GBR), Non-GBR, or Delay-Critical GBR, and are characterized by a set of standard (default) QoS metrics, such as flow Priority Level, end-to-end Packet Delay Budget (PDB), Packet Error Rate (PER), Averaging window (for GBR and Delay-critical GBR resource type), and Maximum Data Burst Volume (for Delay-critical GBR resource type). 5G QoS flows are also characterized by application-specific, operator-defined, or negotiated QoS parameters such as the Guaranteed Flow Bit Rate (GFBR) (for GBR resource type) and Maximum Flow Bit Rate (MFBR) (for GBR resource type). All these parameters are meant to be