5GC-Bench: A Framework for Stress-Testing and Benchmarking 5G Core VNFs

Ioannis Panitsas, Tolga O. Atalay, Dragoslav Stojadinovic, Angelos Stavrou, Leandros TassiulasarXiv 2025 · DOI · arXiv

Introduction

The evolution of 5G mobile networks has transitioned the core architecture from monolithic hardware to a disaggregated, cloud-native Service-Based Architecture (SBA). In this paradigm, network functions are implemented as modular Virtual Network Functions (VNFs) that interact via standardized interfaces. While this shift enhances flexibility and scalability, it introduces significant benchmarking complexities. Traditional tools often evaluate control-plane signaling and user-plane traffic in isolation, failing to capture the interdependencies and resource bottlenecks that emerge under realistic, joint-load conditions.

System overview of 5GC-Bench interacting with the 5G Core and RAN components.

The 5GC-Bench Framework

To address these challenges, we present 5GC-Bench, an open-source framework designed to stress-test 5G Core (5GC) implementations by emulating realistic control and user plane behaviors simultaneously.

5GC-Bench is designed as a modular plugin that interfaces with 5GC implementations such as OpenAirInterface. Its architecture comprises five primary components:

  • Scenario Manager: Allows users to define specific test parameters and stress-test objectives.
  • Orchestrator: Coordinates the execution of defined scenarios across the framework.
  • Control-Plane Load Injector (CPLI): Simulates signaling procedures between User Equipment (UE) and the core network.
  • User-Plane Load Injector (UPLI): Generates and tunnels actual service traffic (e.g., video streaming, gaming) through the network.
  • Telemetry Collector: Monitors system performance at high resolution, capturing CPU, memory, and network metrics via the Docker API.
Internal modular architecture of the 5GC-Bench framework.

Experimental Methodology

The framework was validated on a high-performance testbed utilizing AMD EPYC servers and Ettus N300 software-defined radios. To ensure realistic stress scenarios, the framework leverages three distinct datasets:

  1. Telecom Italia: Models spatial user arrival patterns and signaling activity across 10,000 cells.
  2. NetMob23: Provides service access probabilities (e.g., social media vs. streaming) to reflect real-world usage.
  3. TelecomTS: Contains real-world packet traces for applications like YouTube and Fortnite, enabling realistic traffic replay through the User Plane Function (UPF).

Key Insights and Performance Analysis

Our evaluation revealed critical insights into the resource consumption patterns of various VNFs:

  • Resource Heterogeneity: VNFs exhibit distinct bottleneck characteristics. The Access and Mobility Management Function (AMF) and Session Management Function (SMF) are primarily CPU-bound due to signaling and cryptographic overhead. Conversely, the User Data Repository (UDR) is memory-bound, driven by the need to manage extensive subscriber profiles.
  • Cascading Performance Failures: Due to the chained nature of 5G procedures, a bottleneck in one VNF often triggers performance degradation across others. For instance, a surge in session requests can cause simultaneous CPU spikes across the AMF, SMF, AUSF, UDM, and NRF.
  • User Plane Constraints: The UPF's performance is heavily constrained by CPU cycles required for packet processing, while its memory footprint remains relatively stable regardless of traffic volume, suggesting that scaling the UPF requires additional compute cores rather than RAM.

Results

Experimental results highlight the application-specific impact on core resources. For 500 concurrent sessions, Instagram traffic required significantly more CPU resources (916 millicores) on the UPF compared to online gaming (181 millicores). Furthermore, control-plane functions generally demonstrated linear CPU scaling relative to the number of requests, providing a predictable model for resource planning. In high-load scenarios emulating nearly 2,000 connections, the UPF CPU utilization exceeded 800 millicores, demonstrating the framework's capacity to push systems to their operational limits.

Conclusion

5GC-Bench provides a robust platform for identifying hidden bottlenecks in cloud-native 5G deployments. By jointly emulating control and user plane traffic, it offers a more accurate representation of real-world network stress than isolated synthetic benchmarks. Future work will focus on leveraging these insights to develop automated, proactive scaling policies for 5GC VNFs. The framework is publicly available to support the research community.

Summary updated

Abstract

The disaggregated, cloud-native design of the 5G Core (5GC) enables flexibility and scalability but introduces significant challenges. Control-plane procedures involve complex interactions across multiple Virtual Network Functions (VNFs), while the user plane must sustain diverse and resource-intensive traffic. Existing tools often benchmark these dimensions in isolation, rely on synthetic workloads, or lack visibility into fine-grained resource usage. This paper presents 5GC-Bench, a modular framework for stress-testing the 5GC under realistic workloads. 5GC-Bench jointly emulates signaling and service traffic, supporting both VNF profiling and end-to-end service-chain analysis. By characterizing bottlenecks and resource demands, it provides actionable insights for capacity planning and performance optimization. We integrated 5GC-Bench with the OpenAirInterface (OAI) 5GC and deployed it on a real 5G testbed, demonstrating its ability to uncover resource constraints and expose cross-VNF dependencies under scenarios that mirror operational 5G deployments. To foster reproducibility and further research, we release publicly all the artifacts.
BibTeX
@misc{panitsas2025gc,
  author = {Ioannis Panitsas and Tolga O. Atalay and Dragoslav Stojadinovic and Angelos Stavrou and Leandros Tassiulas},
  title = {{5GC-Bench: A Framework for Stress-Testing and Benchmarking 5G Core VNFs}},
  year = {2025},
  eprint = {2509.18443},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2509.18443},
}
Figure