Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Sub-millisecond Latency and Guaranteed Bit Rates in 5G User Plane

Leonardo Alberro, Noura Limam, Raouf Boutaba

TL;DR

The paper tackles the challenge of delivering end-to-end QoS in 5G transport by proposing a QoS-aware data plane implemented in P4 on an Intel Tofino switch. It maps 3GPP 5QI profiles to four resource types (GBR, GBR*, Non-GBR, Non-GBR*) and combines per-flow metering, classification, strict-priority scheduling, and delay-aware queuing to provide per-flow guarantees and ultra-low latency under congestion. The authors implement the model on real hardware, validate sub-millisecond delays for delay-critical traffic, and demonstrate strong isolation of guaranteed flows with near-zero losses under varying load and high congestion. They also provide a formal framework for delay and throughput predictability and discuss practical hardware constraints and future improvements. Overall, this work offers a programmable, scalable foundation for end-to-end QoS enforcement in future 5G transport networks, enabling precise SLA enforcement and robust performance under dynamic traffic conditions.

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.

Towards Sub-millisecond Latency and Guaranteed Bit Rates in 5G User Plane

TL;DR

The paper tackles the challenge of delivering end-to-end QoS in 5G transport by proposing a QoS-aware data plane implemented in P4 on an Intel Tofino switch. It maps 3GPP 5QI profiles to four resource types (GBR, GBR*, Non-GBR, Non-GBR*) and combines per-flow metering, classification, strict-priority scheduling, and delay-aware queuing to provide per-flow guarantees and ultra-low latency under congestion. The authors implement the model on real hardware, validate sub-millisecond delays for delay-critical traffic, and demonstrate strong isolation of guaranteed flows with near-zero losses under varying load and high congestion. They also provide a formal framework for delay and throughput predictability and discuss practical hardware constraints and future improvements. Overall, this work offers a programmable, scalable foundation for end-to-end QoS enforcement in future 5G transport networks, enabling precise SLA enforcement and robust performance under dynamic traffic conditions.

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.

Paper Structure

This paper contains 26 sections, 11 equations, 8 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. 5G Services and QoS Requirements
  4. Data Plane Programmability with P4
  5. Programmable Data Plane for Enhanced 5G Services
  6. A QoS-aware Transport Network
  7. Programmable Data Plane Model
  8. Design Overview
  9. Classification
  10. Metering and Policing
  11. Scheduling
  12. Control Plane Integration
  13. On the predictability of delay and throughput
  14. Implementation and Evaluation
  15. Evaluation
  16. ...and 11 more sections

Figures (8)

  • Figure 1: A 5G data plane architecture with disaggregated components placed in data centers.
  • Figure 2: 3GPP QoS realization for downlink packets.
  • Figure 3: Programmable pipeline for per-packet QoS processing.
  • Figure 4: Implemented pipeline on the Tofino Native Architecture.
  • Figure 5: Throughput and CDF packet loss using full throughput capacity.
  • ...and 3 more figures