Next Article in Journal
Encrypted Client Hello Is Coming: A View from Passive Measurements
Previous Article in Journal
A Novel Cloud Energy Consumption Heuristic Based on a Network Slicing–Ring Fencing Ratio
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Towards Intelligent 5G Infrastructures: Performance Evaluation of a Novel SDN-Enabled VANET Framework

by
Abiola Ifaloye
,
Haifa Takruri
and
Rabab Al-Zaidi
*
School of Science, Engineering and Environment, University of Salford, 43 Crescent, Salford M5 4WT, UK
*
Author to whom correspondence should be addressed.
Network 2025, 5(3), 28; https://doi.org/10.3390/network5030028
Submission received: 11 June 2025 / Revised: 10 July 2025 / Accepted: 17 July 2025 / Published: 5 August 2025

Abstract

Critical Internet of Things (IoT) data in Fifth Generation Vehicular Ad Hoc Networks (5G VANETs) demands Ultra-Reliable Low-Latency Communication (URLLC) to support mission-critical vehicular applications such as autonomous driving and collision avoidance. Achieving the stringent Quality of Service (QoS) requirements for these applications remains a significant challenge. This paper proposes a novel framework integrating Software-Defined Networking (SDN) and Network Functions Virtualisation (NFV) as embedded functionalities in connected vehicles. A lightweight SDN Controller model, implemented via vehicle on-board computing resources, optimised QoS for communications between connected vehicles and the Next-Generation Node B (gNB), achieving a consistent packet delivery rate of 100%, compared to 81–96% for existing solutions leveraging SDN. Furthermore, a Software-Defined Wide-Area Network (SD-WAN) model deployed at the gNB enabled the efficient management of data, network, identity, and server access. Performance evaluations indicate that SDN and NFV are reliable and scalable technologies for virtualised and distributed 5G VANET infrastructures. Our SDN-based in-vehicle traffic classification model for dynamic resource allocation achieved 100% accuracy, outperforming existing Artificial Intelligence (AI)-based methods with 88–99% accuracy. In addition, a significant increase of 187% in flow rates over time highlights the framework’s decreasing latency, adaptability, and scalability in supporting URLLC class guarantees for critical vehicular services.

1. Introduction

The rapid growth of connected and autonomous vehicles in next-generation networks has increased the need for Ultra-Reliable Low-Latency Communication (URLLC) to support mission-critical services such as autonomous driving, real-time traffic coordination, and collision avoidance within Fifth-Generation Vehicular Ad Hoc Networks (5G VANETs). However, the distributed nature of 5G VANET infrastructures presents significant challenges in managing compute, storage, and radio resources across geographically distributed nodes. Ensuring the efficient offloading and processing of critical vehicular data on edge application servers capable of maintaining optimal Quality of Service (QoS) for time-sensitive applications remains a particularly significant challenge.
Software-Defined Networking (SDN) and Network Functions Virtualisation (NFV) are emerging technologies offering transformative approaches for optimising QoS and resource management in 5G VANETs. Whilst SDN decouples the Radio Access Network (RAN) control and user plane functionalities, to enable a flexible and programmable framework for meeting the diverse and stringent URLLC-class QoS requirements, NFV introduces a new paradigm that allows network functions such as switching, routing, and security to be deployed and scaled on demand to support the increasing IoT and VANET demands in Next-Generation Networks (NGNs) [1].
A key capability of SDN is network slicing, which enables the deployment of Virtualised Network Functions (VNFs) as logically isolated segments over a shared infrastructure. Network slicing configurations may vary. A network slice may include isolated subsets of User Plane (UP) and Control Plane (CP) functions, with dedicated spectrum allocations and cell sites, as seen in Third Generation (3G) and Fourth Generation (4G) networks. Alternatively, slices can be applied across different layers of the protocol stack, shared among heterogeneous services while maintaining isolation [2,3].
SDN’s network slicing capabilities are standardised in the Open Networking Foundation (ONF) SDN architecture, which defines the functional layers, interfaces, and protocols that support dynamic control and virtualisation in Fifth Generation (5G) network environments. The ONF architecture [4] facilitates the efficient management and orchestration of network resources through an abstraction layer that enables a high degree of customisation and responsiveness to meet the stringent URLLC-class QoS requirements in 5G VANETs. In addition, 5G deployment frameworks, architectures, configurations, and use cases are effectively addressed in the Third-Generation Partnership Project (3GPP) release 15 and 16 [1,5].
Recent studies proposing QoS optimisation for connected vehicles in 5G VANETs have explored edge computing frameworks that incorporate both horizontal and vertical offloading strategies to reduce workload to edge and cloud servers [6,7]. In horizontal offloading, vehicles offload data or computational tasks to nearby vehicles via Vehicle-to-Vehicle (V2V) technology, enabling reduced latency. However, this approach increases resource consumption for participating vehicles [6].
In contrast, vertical edge offloading, which is more commonly adopted, shifts data processing from vehicles to edge servers, which offer greater but limited computational capacity. Typically, vertical offloading introduces increased latency and potential network congestion. Therefore, the limited availability of resources and strict latency requirements at the 5G vehicular edge necessitates the application of intelligent management strategies to support vehicles involved in the processing of critical vehicular data, such as Camera, Light Detection and Ranging (LiDAR), and Radio Detection and Ranging (RADAR), and enhance safe offloading process.
The combination of SDN and NFV offers a promising solution to enable predictive workload analysis, dynamic resource allocation, and load balancing across distributed communication infrastructures. These technologies promise to enhance the reliability and availability of vehicular services in 5G VANETs. However, managing virtualised and distributed resources in 5G VANETs remains complex, and recent studies often rely on mathematical models and simulation platforms that do not support 5G implementation frameworks [8,9]. This gap hinders the implementation of QoS and resource management solutions that are tailored to the stringent requirements of critical 5G VANETs applications.
To address this gap, this paper investigates the following research question: How can the integration of SDN and NFV in connected vehicles enhance URLLC-class guarantees and resource management for critical vehicular data in 5G VANETs? We address this through the design and implementation of a novel SDN-Enabled 5G VANET framework that extends computing and storage capabilities to connected vehicles via VNFs, transforming them into transport nodes with embedded network management capabilities. This approach is powered by the open network interfaces of new architectural paradigms like 5G VANETs [10]. The integration of SDN and NFV capabilities in 5G VANETs connected vehicles allows for the pre-processing of critical vehicular data on vehicles to optimise QoS and resource utilisation at the edge. This differs from traditional VANET frameworks, where data processing begins at the Mobile Edge Computing (MEC) layer [11].
The proposed framework leverages SDN principles, including NFV, real-time traffic engineering, and network slicing, to ensure the efficient and reliable management of time-sensitive vehicular applications, promoting safer autonomous driving systems. The contributions of this paper include the following:
  • Novel SDN-Enabled Vehicular Edge Computing Framework: We embed a lightweight SDN gateway and Virtual Extensible Local-Area Network (VXLAN) underlay transport link in every vehicle, turning them into “Software-Defined Vehicles (SDVs)”, capable of local flow classification, slice selection, on-board caching, and reliable data transmission.
  • URLLC-class Guarantees: We demonstrate the ability of the proposed framework to support Ultra-Reliable Low-Latency Communication for the 5G VANET applications responsible for critical vehicular services; this is validated through the 100% packet delivery rate for connected vehicles and 187% increase in flow rates over time.
  • Slice-Aware Resource Management: We propose a resource allocation scheme based on network slicing to enhance radio resource management at the vehicular edge. Our proposed scheme utilises the SDN underlay network to effectively classify critical vehicular data (Camera, LiDAR, and RADAR) into three different QoS classes for dynamic offloading to edge application servers.
The remainder of this paper is organised as follows: Section 2 presents a review of related work. Section 3 outlines the proposed framework and implementation process, including software and hardware requirements, algorithms, and a flow diagram. Section 4 discusses the results, while Section 5 provides a performance evaluation and comparative analysis with existing solutions. Lastly, Section 6 concludes the paper with a summary of the findings, the significance of our study and its limitations, and directions for future work.

2. Related Work

The integration of SDN into distributed 5G infrastructures to optimise QoS and resource management for IoT and VANET applications has been explored in several studies looking at cloud and fog computing solutions, with edge computing solutions becoming more popular in recent years [12]. This section reviews recent advancements in SDN-enabled solutions for QoS and resource management optimisation in 5G VANETs, identifying the key gaps that motivate this study.

2.1. Fifth-Generation Radio Access Network (RAN) Architecture

The 5G RAN architecture supports three main deployment options—enhanced Mobile Broadband (eMBB), massive Machine-Type Communications (mMTCs), and Ultra-Reliable Low-Latency Communications (URLLCs), to meet the diverse IoT application needs in next-generation networks [13]. This includes RAN configurations where Distributed Units (DUs) and Central Units (CUs) are either located together or distributed across the network to enable flexible resource management [3]. Radio Access Network (RAN) slicing, which partitions baseband functions into virtualised segments, enhances scalability and efficient Radio Resource Management (RRM) by granularly allocating resources to specific services in real-time. However, this introduces new network management challenges related to the virtualisation of baseband functions, and the efficient management of these functions on distributed units whilst meeting the required QoS [14]. Traditionally, 5G RAN relies on Next-Generation Node B-Distributed Units (gNB-DUs) for real-time data pre-processing (e.g., filtering and format conversion) to support URLLC [15]. In contrast, our proposed framework extends data pre-processing to connected vehicles in 5G VANETs through the integration of NFV and SDN to enhance QoS and dynamic resource allocation at the gNB for critical vehicular applications.

2.2. SDN and NFV in 5G VANETs

SDN decouples the control and user plane functionalities of the RAN, providing a flexible and programmable framework for managing diverse QoS requirements for URLLC applications in 5G VANETs. The RAN Intelligent Controller (RIC) is an initiative of the Open-RAN (O-RAN) Alliance, combining Extensible RAN (xRAN) and Cloud RAN (C-RAN) to enable more dynamic and intelligent network provisioning and management. This allows for the functional split of the Next Generation Node B (gNB) baseband functions into Real-Time RIC, responsible for dynamic resource allocation, radio admissions control, and RAN data measurements, and Non-Real-Time RIC, which is responsible for inter-cell RRM, resource block control, and connection monitoring [16]. NFV complements SDN through the virtualisation of network functions that traditionally run on dedicated proprietary hardware [1]. NFV allows network functions to be deployed and scaled on demand to support the rapidly growing IoT and VANETs services in next-generation networks.
The integration of an SDN Application Programming Interface (API) Controller into the edge points of an SDN overlay network was explored in [14] in the design of a reliable mesh network via the OmNET modelling system to develop traffic management practices for SDN and SD-WAN communication infrastructures. Their proposed framework achieved a maximum packet delivery rate of approximately 96%, compared to the 83–90% achieved using traditional protocols such as Advanced On-Demand Distance Vector (AODV), Optimised Link State Routing (OLSR), and Better Approach to Mobile Ad Hoc Networking (B.A.T.M.A.N).
The SDN-IoT architecture proposed in [12] includes the integration of IoT devices into the user or data plane of the SDN architecture. Technically, this aligns with our proposed framework, although no experimental results were reported. The SDN-IoT architecture creates a technical change in requirements when designing and implementing communication infrastructures for IoT connectivity on mobile networks. The integration of IoT connectivity into the user plane of the SDN architecture differs from traditional frameworks, with a consolidated structure for communication protocols like the Transmission Control Protocol/Internet Protocol (TCP/IP) and architectures that cannot be disintegrated.
The integration of SDN and NFV into IoT and VANET frameworks faces challenges in supporting dynamic, layered communication architectures. The study conducted in [9] highlights the lack of comprehensive frameworks featuring new layers in communication infrastructures and protocols. Although their proposed IoTNetSim framework allows for the end-to-end simulation of IoT services, including device interactions and communication protocols like Message Queueing Telemetry Transport (MQTT), it lacks support for SDN integration. This is a significant gap as SDN’s centralised control is essential for optimising QoS in 5G VANETs.

2.3. Network Slicing and Machine Learning for Resource Management

SDN-enabled solutions for enhancing network flexibility and scalability in next-generation IoT and VANETs include Machine Learning (ML) frameworks, that utilise the SDN Controller API for centralised resource management. The comprehensive survey conducted in [17] effectively captured the various applications of Artificial Intelligence (AI) and ML in next-generation networks, including techniques for traffic and load prediction, congestion control, traffic flow, QoS, and resource management. Network slicing is the key SDN principle driving innovation for SDN and AI-based solutions, enabling the deployment of Virtualised Network Functions (VNFs) for enhanced network scalability and flexibility. A network slice is a customised portion of a network, deployed to fulfil the specific requirements of a particular use case. This provides an avenue to deploy several logical segments that may be isolated on a shared infrastructure.
Flow-prioritisation techniques were applied in [18] to classify traffic flows, via the SDN controller, into bandwidth-intensive and priority queues to enable dynamic resource allocation, achieving an accuracy of approximately 93%, 95%, and 99% for the deep learning models Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM), respectively. LSTM and Q-Learning algorithms were also applied in [19] to address the resource allocation decision problem in VANETs, with a maximum accuracy of approximately 88%. Similarly, a Multi-Layer Perception (MLP) neural network was applied in [7] to predict offloading success across a Wireless Local Network (WLAN) 802.11 p, and RSU (Road-Side Unit) and cellular Long-Term Evolution (LTE) networks. The MLP classifier achieved an accuracy of 81% for edge offloading due to the overutilisation of edge servers.

2.4. Research Gap

A significant gap in the literature is the integration of IoT connectivity into the user plane of the SDN architecture [12]. Whilst recent studies have focused on SDN optimisation for 5G VANETs via the control plane [14], our proposed framework explores optimisation via the user plane, achieving a consistent packet delivery rate of 100% for connected vehicles. Furthermore, whilst AI frameworks integrating ML into SDN Controller APIs offer greater flexibility in resource management, these frameworks strongly rely on large datasets and reliable models capable of guaranteeing the URLLC-class, particularly for critical vehicular services. To ensure the URLLC-class guarantees for critical vehicular services, it is imperative that operations such as real-time traffic classification are 100% accurate, as demonstrated in our study.

3. Proposed Method

This section describes our proposed framework for critical vehicular services in 5G VANETs. To address the unique challenges for ensuring URLLC-class guarantees, a novel SD-WAN framework that integrates NFV, SDN, and edge computing into connected vehicles was designed to enhance communication reliability and scalability for critical 5G VANET services. The following subsections detail the proposed framework, including the research methodology, architectural design, implementation framework, and operational mechanisms.

3.1. Research Methodology

We followed a scientific and iterative method, beginning with a comprehensive literature review and problem formulation based on gaps identified in the literature. A theoretical analysis of architectural designs, models, and protocols informed the identification of system requirements. These requirements were then converted into a system model, including the performance and user specifications relevant to critical 5G VANETs applications.
The next phase involved the implementation and development of a prototype system using Agile Model-Driven Development (AMDD) methodology, which allowed for the progressive refinement of both the design and implementation through iterative testing. Following implementation, an empirical analysis was conducted and data collected to evaluate current system performance.
The collected results were also used for iterative refinement, ensuring that the technical objectives were met. When refinements were not satisfactorily met, insights from previous iterative tests were used to reconfigure and improve the design. Once the desired benchmarks were met, we proceeded to the final phase, which included a comprehensive analysis of the results and the documentation of all processes and procedures undertaken.

3.2. Proposed Architecture

The proposed architecture (Figure 1) extends the computing functionalities usually available in remote edge and cloud datacentres in traditional VANETs architecture [11] to connected vehicles in 5G VANETs. The open network interfaces of new network paradigms like 5G VANETs enable vehicles to act as edge transport nodes with embedded network management functionalities [10].
In contrast to the traditional VANETs architecture, vehicles are SDN-enabled in the proposed architecture. Although three main classes of IoT communications, enhanced Mobile Broadband (eMBB), massive Machine-Type Communication (mMTC), and URLLC, are found in 5G networks [13], the proposed architecture is specifically directed towards URLLC in 5G VANETs. URLLC or the critical IoT data under this category are time-sensitive and require ultra-reliable communication. In addition, the diverse nature of the traffic characteristics and sensor patterns under this category requires further classification to satisfy their unique requirements.
The proposed architecture integrates NFV and SDN into connected vehicles in 5G VANETs to optimise QoS for the Camera, LiDAR, and RADAR sensors required for autonomous driving and other emergency scenarios, like collision avoidance. Instead of transmitting vehicular data directly to the gNB, all data generated by vehicular sensors are forwarded to a local SDN gateway embedded within the vehicle’s On-Board Unit (OBU).
The SDN gateway (Figure 2) is responsible for the initial management of vehicular sensor data using a pre-installed SDN controller algorithm. This includes packet inspection, prioritisation, and the preparation of data for further processing at the gNB. The SDN Controller algorithm ensures that packets are efficiently processed using specified QoS and forwarding logic; these are unique to the critical vehicular sensor data that are currently handled being on the network. Whilst the SDN gateway is responsible for the flow of data in the user plane, our proactive SDN controller algorithm centrally orchestrates and manages the control plane of the SDN-Enabled Vehicular Network.
The majority of research in this field has explored QoS and resource management optimisation in 5G VANETs by placing edge datacentres at the gNB. Network slicing is at the heart of these designs, optimising QoS and resource management via the integration of NFV and SDN at gNB. The most common SDN solutions employ OpenFlow protocol at the gNB to proactively and reactively manage network operations. To extend the coverage of the edge and cloud datacentres, SDN overlay protocols like VXLAN and Internet Protocol Security (IPSEC) are often used [14].
The extension of VXLAN technology to connected vehicles in 5G VANETs, as demonstrated in this study, offers better scalability and reliability. The implementation of fully virtualised, and distributed 5G VANET architectures involves the joint management of the VNFs deployed across various geographical locations via matching implementation policies at each location. The proposed architecture applies a novel SDN Controller algorithm via a VXLAN underlay network to enable seamless communication between VNFs hosted with SDN-Enabled or Software-Defined Vehicles (SDVs) and application servers located at the gNB.

3.3. Implementation Framework

This section describes the implementation of the proposed architecture for vehicular data (Camera, LiDAR, and RADAR) in 5G VANETs. To implement the functional requirements, four Ubuntu Virtual Machines (VMs) with two cores and 2 GB RAM were deployed as SDVs. At the gNB, one Ubuntu VM with four cores and 2 GB RAM was deployed to host the edge datacentre.
The implementation of the proposed architecture followed a software-defined approach. In software-defined networking, network applications are deployed to manage network functions. A novel event-driven SD-WAN framework (Figure 3) was deployed using the available resources on VMs to orchestrate and manage the flow of communication between SDVs and vehicular application servers hosted by the gNB. As illustrated in Figure 3, the data plane performs packet forwarding and traffic classification, and assigns Slice IDs (an identifier used to classify data flows based on QoS class requirements) via OpenFlow-enabled switches. The control plane manages dynamic flow decisions and routing logic, enabling the programmable handling of vehicular communication through SDN Controllers. Additionally, the application plane hosts the SDN control logic responsible for generating flow rules, monitoring link conditions, and orchestrating real-time access management policies.
The virtual network was simulated using Mininet (v2.3) and Python (v3.8) APIs to create a topology linking each SDV to the gNB using a dedicated VXLAN transport tunnel. A star topology was adopted, where each SDV serves as a hub for ten connected sensors that communicate with an embedded SDN gateway for local data pre-processing before transmission to the gNB. On each SDV, we simulated six Camera sensors transmitting approximately 105,260 bytes/second, three LiDAR sensors transmitting 2632 bytes/second, and two RADAR sensors transmitting 10.5 bytes/second to the SDN gateway. The decision regarding the amount of data generated by each sensor was informed by [20], although a significantly lower ratio was implemented due to the testbed capacity.
The proposed framework operates via a joint controller algorithm deployed on SDVs and at the gNB. Upon connection, the SDN Controller installs flow rules on OpenFlow tables 0 and 1 at the SDN gateway and gNB. Figure 4 illustrates the end-to-end flow of vehicular data from collection to offloading, which is aligned with the SDN architecture and Open Systems Internetworking (OSI) model. The flow diagram is divided into two main blocks—Data Collection and Pre-Processing at the SDN gateway, and gNB Processing and Data Offloading at the edge servers. Each block highlights the interplay between the data plane (OSI Layer 2) and the control plane (OSI Layer 3) in orchestrating packet flow and ensuring adequate data handling through OpenFlow tables.
At the SDN gateway, incoming vehicular data is first verified for authenticity using the pre-assigned network ID in OpenFlow table 0 at the data plane. If valid, the data proceeds to Table 1, where port matching and Slice ID assignment are performed within the data plane. The control plane monitors these processes and provides the decision logic used to verify the network parameters. Once a valid Slice ID is assigned, the packet is forwarded through a VXLAN link to the gNB. The VXLAN operation simulates underlay Layer 2 transport tunnelling, which is essential for virtualising the network links within the SDN architecture.
Upon arrival at the gNB, data is routed into OpenFlow table 0, where the VXLAN input port is validated at the data plane. The control plane logic verifies the corresponding port ID. If successful, the data moves to OpenFlow table 1, where Slice ID verification and offloading decisions are made. Finally, based on the Slice ID, the system dynamically offloads the data to its corresponding application server—Camera, LiDAR, or RADAR—and stores the network packet information on databases via the SDN application plane.

3.4. Algorithms

The process begins after a wireless connection is established between SDVs and the gNB. The SDN Controller (Ryu Controller) installs the flow rules to be used by the SDN gateway (Open vSwitch) and the gNB (Open vSwitch) to orchestrate and manage communication flow. The SDN Controller algorithms are implemented at the SDN gateway that resides on vehicles and the gNB to enable real-time data processing, network slicing, and dynamic offloading to edge application servers.

3.4.1. SDN Gateway Matching Functions

The SDN gateway pre-processes vehicular data using two OpenFlow tables Tv(0) and Tv(1), with corresponding matching rule functions Rv(0) (Algorithm 1) and Rv(1) (Algorithm 2). The input ports are defined as Pv(in) = {1, 2,…,10}, and the output port is Pv(out) = 13. The sensor data mappings are Camera (Ports 1–5), LiDAR (Ports 6–8), and RADAR (Ports 9 and 10).
Algorithm 1.Vehicle Matching Rule Function Rv(0) on Table Tv(0)
input:D   //Vehicular Data
x   //Network ID of data
output:D forwarded to Table Tv(1)
//Step 1:Process incoming data
1if x = “0.0.0.0” then
2     forward D to Table Tv(1)
3else
4     drop D
5end if
Algorithm 2.Vehicle Matching Rule Function Rv(1) on Table Tv(1)
input:D       //Vehicular Data
Pv(in)   //Input port
d       //Data type (Camera, LiDAR, RADAR)
output:D with assigned Slice ID forwarded to Pv(out) = 13
//Step 1:Match input data to appropriate class
1if Pv(in) ∈ {1, 2, 3, 4, 5} and d = “Camera” then
2     Slice ID ← 1
3else if Pv(in) ∈ {6, 7, 8} and d = “LiDAR” then
4     Slice ID ← 2
5else if Pv(in) ∈ {9, 10} and d = “RADAR” then
6     Slice ID ← 3
7end if
//Step 2:Forward to output port
8forward D with Slice ID to Pv(out)
All SDVs were configured to communicate directly with the gNB using VXLAN Port 13. Upon the receipt of vehicular data on SDN gateway’s table 0, Tv(0), a packet inspection is performed to extract the Network ID of the received vehicular data using the SDN gateway’s matching rule function 0 Rv(0). The retrieved Network ID is used to verify the integrity of incoming vehicular data before forwarding it to SDN gateway’s table 1, Tv(1). On Tv(1), matching rule function 1 Rv(1) is implemented to match the input port Pv(in) and data type (d) of received vehicular data (D) before a unique Slice ID is assigned.

3.4.2. gNB Matching Functions

At the gNB, two OpenFlow tables Tg(0) and Tg(1), process vehicular data using matching rule functions Rg(0) (Algorithm 3) and Rg(1) (Algorithm 4). The input ports are Pg(in) = {1, 2,…, n}, and the output ports are Pcam, Plid, and Prad for Camera, LiDAR, and RADAR data, respectively. The gNB matching functions verify the integrity of the received vehicular data and dynamically offload to edge application servers using the vehicle-assigned Slice ID.
Algorithm 3.gNB Matching Rule Function Rg(0) on Table Tg(0)
input:D with Slice ID     //Vehicular Data with Slice ID
Pg(in)            //Input port from SDV to gNB
output:D forwarded to Table Tg(1) or dropped
//Step 1:Verify source of incoming data
1if Pg(in) ∈ {1, 2, …, n} then
2     forward D to Table Tg(1)
3else
4     drop D
5end if
Algorithm 4.gNB Matching Rule Function Rg(1) on Table Tg(1)
input:D with Slice ID       //Vehicular Data with Slice ID
output:D forwarded to corresponding output port (Pcam, Plid, Prad)
//Step 1:Match data by assigned Slice ID
1if Slice ID = 1 then
2     forward D to Pcam
3else if Slice ID = 2 then
4     forward D to Plid
5else if Slice ID = 3 then
6     forward D to Prad
7end if
The proposed SDN Controller functions presented above apply different SDN principles in the reception, processing, and offloading of vehicular data to edge application servers. On SDVs, NFV was applied to create an IoT network connected to a local SDN gateway via VNFs. The SDN Controller’s matching functions enable the real-time management of vehicular data in queues via OpenFlow tables. Real-time data analytics, traffic engineering, and dynamic SDN forwarding principles were applied in the data plane (Underlay network) of the SDN architecture. Lastly, the SDN Controller matching functions at the SDN gateway and gNB enable end-to-end network slicing for service differentiation and the dynamic offloading of critical vehicular data to edge application servers.

4. Results and Discussion

In this section, we present the empirical results from the implementation of the proposed SDN-enabled VANET framework. To evaluate the functional requirements, we examined the system’s ability to locally classify vehicular sensor data using the computing and storage resources available in SDVs and to dynamically assign vehicular requests to the appropriate servers at the gNB. To evaluate the non-functional requirements, we analysed key performance indicators, such as network delay, packet loss, flow rates, and flow durations, collected via the SDN application plane. These metrics provide insight into the system’s operational efficiency and reliability.
We adopted a two-phase testing strategy. In the first phase, a Unit testing was conducted to evaluate the functional performance of individual components. This included testing of data integrity, traffic classification, communication flow, and resource allocation. This was followed by integration testing, where we examined interactions between modules and their coordination in processing and transmitting vehicular data.
During preliminary testing, we observed a limitation in the simulation environment (Mininet), which restricted continuous data transmission to approximately 15 min. To accommodate this challenge, we structured each simulation session to include three separate transmission tests, amounting to 40 min per session. In total, ten simulation sessions were conducted, yielding 30 test runs. The aggregated results from these sessions were analysed to assess the overall performance of the framework across key metrics: traffic classification accuracy, transmission success, flow duration, flow rate, delay, and packet loss.

4.1. Traffic Classification and Transmission

The framework achieved a 100% in-vehicle traffic classification accuracy, indicating the efficient differentiation of critical vehicular data (Camera, LiDAR, and RADAR) into three distinct classes using Slice IDs (Figure 5). In 5G VANETs, this classification is useful for identity management, service differentiation, and QoS prioritisation. Complementing this result, packet transmission analysis (Figure 6) showed that all data transmitted by the SDVs were successfully received at the gNB with zero packet loss, demonstrating the reliability of the SDN VXLAN underlay transport network.
During preliminary testing, we observed that high transmission bursts from Camera sensors, due to their larger packet sizes and higher data rates, increased flow durations in processing units and led to approximately 50% packet loss. To address this, we conducted a pre-simulation analysis and profiling of sensor transmission characteristics, including packet sizes, current flow durations, and transmission rates. This informed the tuning of transmission parameters and configurations, enabling the system to achieve 100% transmission reliability in subsequent tests.

4.2. Flow Duration

Flow duration analysis plays a crucial role in evaluating the real-time communication performance within 5G infrastructures, particularly given the distributed nature of processing units at vehicular edge nodes. In our implementation, flow duration served as an important metric for assessing the SDN Controller’s efficiency in managing vehicular traffic while maintaining the required QoS. Because we modelled the system manually to simulate the behaviour of real-world sensors, measuring flow duration helped to validate traffic flow patterns and detect bottlenecks during live data transmission.
As illustrated in Figure 7, Camera and LiDAR sensors consistently recorded shorter flow durations at the SDN gateway compared to RADAR sensors. This pattern remained stable across all three test phases, with average flow durations decreasing over time for all sensor types. Specifically, flow durations for Camera and LiDAR dropped from 840 s in Test 1 to approximately 292 s in Test 3, while those of RADAR decreased from 1221 s to 406 s. These improvements suggest increasing efficiency in the SDN Controller’s scheduling and pre-processing logic.
Notably, despite transmitting the smallest volume of data, RADAR sensors experienced the longest flow durations. This was due to their lower packet sizes and transmission rates, which increased the queueing and wait times at processing units. In latency-sensitive 5G VANET environments, such delays could compromise real-time applications, such as obstacle detection or braking response, underscoring the importance of prioritised scheduling for RADAR sensors in future implementations.

4.3. Flow Rate

In 5G VANETs, where ultra-reliable and low-latency communication is critical, understanding flow rate dynamics is useful for optimising network performance and maintaining consistent data delivery. In our implementation, flow rate was closely monitored at both the SDN gateway and the gNB to assess the system’s adaptability to varying vehicular loads whilst maintaining dynamic QoS management.
At the SDN gateway, we implemented rate-limiting to structurally manage the transmission of data from Camera, LiDAR, and RADAR sensors. This was particularly important given the constrained computation and network resources at the edge. As illustrated in Figure 8, Camera sensors achieved the highest flow rate, increasing from an average of 2.3 packets per second (PPS) in Test 1 to 6.6 PPS in Test 3, an approximate increase of 187%. LiDAR and RADAR sensors also recorded gains, rising from 1.4 PPS to 4 PPS and 0.6 PPS to 1.9 PPS, respectively. These results demonstrate the system’s scalability and its ability to maintain reliable throughputs under increasing load, especially for high-bandwidth sensors such as cameras.
  • Complementing this, flow rate analysis at the gNB further confirmed the system’s ability to support high-throughput offloading. These flow rates represent not only the receipt of pre-processed data from the SDN gateway but also the rate at which Slice IDs were retrieved and subsequently used to dynamically route vehicular traffic to edge application servers. As illustrated in Figure 9, Camera sensor rate at the gNB increased from 8.4 PPS in Test 1 to 18 PPS in Test 3, an increase of approximately 115%. LiDAR and RADAR sensors followed a similar upward trend, with flow rates rising from 4.3 PPS to 11 PPS and 2.4 PPS to 5.9 PPS, respectively.
The consistent increase in flow rates across the system reflects the robustness and adaptability of the SDN Controller algorithm in managing critical vehicular data. This scalability is particularly important for URLLC applications in 5G VANETs to maintain high throughput and stable transmission for autonomous driving, collision avoidance, and other mission-critical services.

4.4. Delay

Delay is a critical performance metric for 5G VANETs, particularly when real-time responsiveness is essential for mission-critical applications. In our analysis, we measured average delay per-packet at both the SDN gateway and the gNB across all three test phases to assess how efficiently the framework processed and offloaded critical vehicular data to edge application servers.
As illustrated in Figure 10, the SDN gateway exhibited significant improvements in delay performance over time. Camera sensor delay dropped from 430 milliseconds (ms) in Test 1 to 150 ms in Test 3, a reduction of approximately 65%. LiDAR delay dropped from 720 ms to 250 ms, while RADAR dropped from 1570 ms to 500 ms. These results reflect the growing efficiency of the SDN Controller in scheduling and pre-processing data streams as traffic volume increased over time.
Despite these improvements, the delay at the SDN gateway remained higher than that at the gNB, mainly due to the overhead involved in local pre-processing and slice selection. As illustrated in Figure 11, the gNB consistently recorded lower delay values during the offloading phase. In Test 3, where the overall best results were observed, average delays at the gNB were 60 ms for Camera, 90 ms for LiDAR, and 170 ms for RADAR. These delay values are significantly lower than those at the SDN gateway, indicating better data handling once traffic is classified and transmitted upstream.
When considering the total end-to-end delay, including in-vehicle pre-processing and gNB offloading, the system achieved a combined delay of 210 ms for Camera, 340 ms for LiDAR, and 670 ms for RADAR during the third test. These figures demonstrate improved performance compared to successive tests, although end-to-end delay needs further optimisation.

4.5. Packet Loss

Whilst the proposed framework achieved 100% successful data transmission from connected vehicles to the gNB, supported by VXLAN virtualisation of the Radio Access Network, packet loss was observed during the final offloading to edge application servers. An average packet loss of approximately 0.38% for Camera, 2.7% for LiDAR, and 7% for RADAR was recorded despite the successful retrieval of all pre-assigned Slice IDs at the gNB. This performance is attributed to the non-configuration of VXLAN transport network between the gNB and edge application servers, highlighting an implementation gap.
In summary, the proposed SDN-Enabled 5G VANET framework demonstrates strong performance across key metrics: 100% classification accuracy, zero packet loss between vehicles and the gNB, a 187% improvement in flow rates over time, and an up to 68% reduction in delay. Nevertheless, limitations remain, particularly in packet handling during offloading to edge servers and latency at the SDN gateway. Additionally, the current system simulates a 5G infrastructure over a 4G network, which may inherently introduce overhead delays that would not be expected in native 5G deployment.

5. Performance Evaluation and Comparative Analysis

In 5G VANETs, where network resources including hardware and software are distributed across different geographical locations, it is imperative to match vehicular requests to the right resources that can guarantee the required QoS. Furthermore, in a fully virtualised 5G communication infrastructure for critical VANETs applications, where network resources including the virtual Radio Access Network (vRAN) are collectively managed by the SDN Controller, the granular matching of vehicular requests to network functions presents novel opportunities for ensuring optimal QoS for critical operations. The disintegration of the application, control, and user planes in SDN enables the seamless matching of network functions across the three planes.
To evaluate the performance of the proposed framework, we applied queueing and delay theories to queue vehicular data (events) across a chain of distributed and virtualised network resources (queues), including communication links. The proposed event-driven multi-queue and delay broker framework (Figure 12) employs a broker (SDN controller flow rules, R1-RN) to process vehicular data on a chain of queues (VNFs) including the SDN gateway, VXLAN links to gNB, gNB, and links to edge application servers.
Traditionally, VANET frameworks rely on layered communication architectures that apply a fixed set of rules at each layer. This rigid structure limits scalability and dynamic access, making it difficult to achieve optimal QoS in 5G VANETs. In contrast, event-driven software architectures are distributed and asynchronous, offering high scalability and adaptability to single-purpose event-processing components [21]. Event-driven software models consist of highly decoupled processing units that enable the granular and strategic management of processing units using pre-defined sets of rules.

5.1. Performance Evaluation

We considered M/M/1/K queueing systems where the arrival and services processes are (M)emoryless, there is only one server within a single queue, and K represents the system capacity. Each queue is modelled as a single M/M/1 queue with K capacity. The processing of critical vehicular data, including Camera, LiDAR, and RADAR, with similar QoS requirements but varying transmission rates and packet sizes, creates the problem of service differentiation to ensure optimal QoS. The proposed model follows the FIFO (First-In-First-Out) packet processing methodology.
From our analysis, the management of diverse critical vehicular data on distributed and shared resources creates the problem of longer delays, particularly for sensors with lower transmission rates and packet sizes, like RADAR. Considering the observed results in Table 1, where the arrival rates are consistently equivalent to the service rates observed at the SDN gateway and VXLAN link to gNB.
ρ = λ / μ     1
The queue-utilisation function ( ρ ) is approximately equal to 1 across the three tests. This implies that the SDN gateway and VXLAN link to gNB were operating at full utilisation across the three tests. The arrival rate at the gNB is the rate on VXLAN links. Therefore, the relatively higher service rates observed at the gNB (as illustrated in Table 2) indicate ρ < 1 at the gNB for all test scenarios.
At the gNB, the system utilisation increased by a small margin between test 1 and test 3 (as illustrated in Table 3), possibly due to increasing arrival rates. However, the maximum utilisation ( ρ ) of approximately 36% suggests that the gNB is far from full utilisation. For queueing systems with finite capacity, K is the maximum number of packets that can be in the system. Calculating the average number of packets in the system and the waiting time often involves state probabilities and considering the probability of packet loss due to full utilisation. Assuming a K of 1940, 1169, and 780 for Camera, LiDAR, and RADAR, respectively, during the third test (as illustrated in Table 4), we can calculate the probability of the system being full P(K) when ρ = 1 as follows:
P K = 1 K + 1
where K is the number of packets the system can hold, excluding the one currently being served. The probabilities that the system is full for the Camera, LiDAR, and RADAR queues formed in the system are therefore approximately 0.0005%, 0.0009%, and 0.0013%, respectively, during the third test. These low probabilities imply that even though the SDN gateway and VXLAN links currently operate at near full utility, it is unlikely that their performance will deteriorate under the current system dynamics.
Taking into consideration the above low probabilities and the 100% packet delivery rate observed across all tests, we can assume that the current capacity K is the average number of packets processed in a queue (Lq). Therefore, the average waiting time for a packet in a queue (Wq) is as follows:
W q   =   L q / λ
Assume Lq is 1940, 1169, and 780 for the arrival rates ( λ ) of 6.6, 4.0, and 1.9, respectively, observed for Camera, LiDAR, and RADAR sensors at the SDN gateway during the third test. Wq is approximately 294, 292.3, and 410.5 s, respectively (as illustrated in Table 5). This is the total time taken (average flow duration) by the SDN gateway to pre-process and transmit all vehicular traffic to the gNB for dynamic offloading.
Using the equation below, the average delay (Li) at the SDN gateway is approximately 0.15, 0.25, and 0.5 s/packet, respectively, during the third test. This implies an average delay per packet of approximately 150, 250, and 500 milliseconds, respectively, for Camera, LiDAR, and RADAR sensors at the SDN gateway (as illustrated in Table 6). This corresponds with the observed delay per packet observed during our simulations.
L i = W q / L q
Applying the same formula and considering the average number of packets offloaded at the gNB for all four vehicles (as illustrated in Table 7) and the average delay observed at gNB during offloading (as illustrated in Table 8), the average delay per packet (Li) at the gNB during the third test is approximately 60, 90, and 170 milliseconds for Camera, LiDAR, and RADAR sensors, respectively (as illustrated in Table 9).
The overall delay per packet for pre-processing vehicular traffic at the SDN gateway, using vehicle on-board resources and dynamically offloading to edge application servers during the third test phase, where the best overall performance was observed, is approximately 210, 340, and 670 milliseconds for Camera, LiDAR, and RADAR, respectively (as illustrated in Table 10).

5.2. Comparative Analysis

The proposed framework demonstrated superior performance across multiple metrics when compared to existing frameworks in the literature. Notably, it explored interference mitigation via the SDN underlay network, which aligns with the theoretical framework presented in [12]. In comparison to the SDN overlay model proposed in [14], which achieved a maximum packet delivery rate of approximately 96% and dropped to as low as 81% under some test conditions, our framework consistently achieved 100% packet delivery rate across all tests (Figure 13). This confirms its enhanced reliability. In terms of resource management, the proposed framework also performed better than the RAN slicing model proposed in [22], which supported only two classes of IoT—massive IoT and URLLC. Unlike [22], our framework provides more granular classification for critical vehicular data types within the URLLC class. At an average flow rate of 2.3, 1.4, and 0.6 PPS for Camera, LiDAR, and RADAR sensors, respectively, our framework maintained a system utilisation of approximately 1; in contrast, they reported less efficient system utilisation values of 1.3, 1.4, and 1.5 at the same flow rates. In addition, our in-vehicle traffic classification model achieved 100% accuracy, outperforming the in-network ML classifier proposed in [23], which reported an average accuracy of 93.5%. Furthermore, it exceeded Deep Learning (DL) models, Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Deep Neural Network (DNN) when applied for flow-based classification in [18] (Figure 14). Whilst AI models offer significant flexibility for dynamic network management in next-generation networks, we recommend that such approaches are reserved for complex, multi-factor resource management tasks. Fundamental network operations, such as traffic classification, can be handled more efficiently using pro-active SDN controller algorithms (as demonstrated in this study) or direct configuration on OpenFlow switches.

6. Conclusions

This study focused on the development of a reliable, adaptive, and cooperative SDN-Enabled 5G VANET framework designed to dynamically adjust to network conditions to enhance dynamic QoS management for connected vehicles. In contrast to existing solutions, the proposed framework leveraged an SDN underlay network to achieve 100% packet delivery rate across all tests, demonstrating significant reliability in vehicular communications. As highlighted in the survey presented in [24], there remains a gap in its practical implementations regarding the need to adequately address the complexities of deploying SDN-Enabled 5G Infrastructures, including routing-based interference mitigation techniques. This paper contributes a functional framework and practical insight into the application of SDN principles with next-generation IoT and vehicular networks. Performance evaluations showed that integrating SDN Controller algorithms for real-time traffic and resource management supports the development of scalable and reliable solutions that support the stringent QoS demands of critical vehicular applications. One limitation observed from our study is the introduction of processing delays at the SDN gateway, which requires further optimisation. In addition, this paper focused on reporting the preliminary performance evaluation of the proposed framework and did not include comprehensive capacity-testing on a 5G network. Future work will explore these limitations, alongside other key challenges such as energy efficiency for resource-constrained edge nodes, ultra-low-latency demands, and the impact of high vehicular mobility in 5G VANET environments.

Author Contributions

Conceptualization, A.I.; software, A.I.; validation, A.I.; investigation, A.I.; resources, H.T. and R.A.-Z.; data curation, A.I. and R.A.-Z.; writing—original draft, A.I.; writing—review and editing, H.T. and R.A.-Z.; visualisation, A.I.; supervision, H.T. and R.A.-Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

The data used to support the findings of this study are included within the article.

Acknowledgments

The authors acknowledge the support of the University of Salford Computer Networking and Telecommunications Research Centre for providing institutional resources and facilities that contributed to the successful completion and publication of this research.

Conflicts of Interest

The authors declare no conflicts of interest.

Abbreviations

The following abbreviations are used in this manuscript:
3GPPThird-Generation Partnership Project
5GFifth-Generation Mobile Networks
5G xHAUL5G Transport Network
AIArtificial Intelligence
APApplication Plane
APIsApplication Programming Interfaces
CNNConvolutional Neural Network
C-RANCloud Radio Access Network
CPControl Plane
CUCentral Unit
DLDeep Learning
DNNDeep Neural Network
DPData Plane
DUDistributed Unit
eMBBEnhanced Mobile Broadband
gNBNext-Generation Node B
IoTInternet of Things
LiDARLight Detection and Ranging
LSTMLong Short-Term Memory
MLMachine Learning
MLPMulti-Layer Perception
mMTCMassive Machine-Type Communications
NFVNetwork Functions Virtualization
NGNNext-Generation Network
ONFOpen Networking Foundation
O-RANOpen Radio Access Network
OBUOn-Board Unit
OSOperating System
OSIOpen Systems Internetworking
QoSQuality of Service
RADARRadio Detection and Ranging
RANRadio Access Network
RICRAN Intelligent Controller
RSURoad-Side Unit
SD-WANSoftware-Defined Wide Area Network
SDNSoftware-Defined Networking
UEUser Equipment
UPUser Plane
URLLCUltra-Reliable Low-Latency Communication
V2VVehicle-to-Vehicle
VANETsVehicular Ad Hoc Network
vCPEVirtual Private Customer Premises Equipment
VMVirtual Machine
VNFVirtual Network Function
VXLANVirtual eXtensible Local-Area Network

References

  1. ETSI. 5G Management and Orchestratio—Architecture Framework (Release 15). ETSI TS 128 533 V15.00.00, October 2018. Available online: https://www.etsi.org/deliver/etsi_ts/128500_128599/128533/15.00.00_60/ts_128533v150000p.pdf (accessed on 21 February 2024).
  2. Schmidt, R.; Chang, C.Y.; Nikaein, N. Slice Scheduling with QoS-Guarantee towards 5G. In Proceedings of the IEEE Global Communications Conference (GLOBECOM), Waikoloa, HI, USA, 9–13 December 2019; pp. 1–7. [Google Scholar]
  3. Adamuz-Hinojosa, O.; Muñoz, P.; Ameigeiras, P.; Lopez-Soler, J.M. Sharing gNB Components in RAN Slicing: A Perspective from 3GPP/NFV Standards. In Proceedings of the IEEE Conference on Standards for Communications and Networking (CSCN), Tokyo, Japan, 2–4 October 2019; pp. 1–7. [Google Scholar]
  4. ETSI. Network Functions Virtualisation (NFV) Release 3; Evolution and Ecosystem; Report on Network Slicing Support with ETSI NFV Architecture Framework. ETSI GR NFV-EVE 012 V3.1.1, December 2017. Available online: https://www.etsi.org/deliver/etsi_gr/NFV-EVE/001_099/012/03.01.01_60/gr_nfv-eve012v030101p.pdf (accessed on 11 May 2021).
  5. ETSI. 5G.; System Architecture for the 5G System; Overall Description; Stage 2. ETSI TS 122 186 V16.02.00, May 2020. Available online: https://www.etsi.org/deliver/etsi_ts/122100_122199/122186/16.02.00_60/ts_122186v160200p.pdf (accessed on 21 February 2024).
  6. Higuchi, T.; Rabsatt, R.V.; Gerla, M.; Altintas, O.; Dressler, F. Cooperative Downloading in Vehicular Heterogeneous Networks at the Edge. In Proceedings of the IEEE Globecom Workshops (GC Wkshps), Waikoloa, HI, USA, 9–13 December 2019; pp. 1–5. [Google Scholar]
  7. Sonmez, C.; Tunca, C.; Ozgovde, A.; Ersoy, C. Machine Learning-based Workload Orchestrator for Vehicular Edge Computing. IEEE Trans. Intell. Transp. Syst. 2021, 22, 2239–2251. [Google Scholar] [CrossRef]
  8. Hu, S.; Li, G. Dynamic Request Scheduling Optimisation in Mobile Edge Computing for IoT Applications. IEEE Internet Things J. 2020, 7, 1426–1437. [Google Scholar] [CrossRef]
  9. Salama, M.; Elkhatib, Y.; Blair, G. IoTNetSim: A Modelling and Simulation Platform for End-to-End IoT Services and Networking. In Proceedings of the 12th IEEE/ACM International Conference on Utility and Cloud Computing, Cambridge, UK, 9–12 December 2019; pp. 251–261. [Google Scholar]
  10. Mur, D.C.; Flegkas, P.; Syrivelis, D.; Wei, Q.; Gutierrez, J. 5G-XHaul: Enabling Scalable Virtualisation for Future 5G Transport Networks. In Proceedings of the 15th International Conference on Ubiquitous Computing and Communications and 2016 International Symposium on Cyberspace and Security (IUCC-CSS), 14–16 December 2016; pp. 173–180. [Google Scholar]
  11. Liu, L.; Chen, C.; Pei, Q.; Maharjan, S.; Zhang, Y. Vehicular Edge Computing and Networking: A Survey. Mob. Netw. Appl. 2021, 26, 1145–1168. [Google Scholar] [CrossRef]
  12. Jazaeri, S.S.; Jabbehdari, S.; Asghari, P.; Haj Seyyed Javadi, H. Edge Computing in SDN-IoT Networks: A Systematic Review of Issues, Challenges, and Solutions. Clust. Comput. 2021, 24, 3187–3228. [Google Scholar] [CrossRef]
  13. Ericsson. Cellular IoT in the 5G Era. White Paper, February 2020. Available online: https://www.ericsson.com/en/reports-and-papers/white-papers/cellular-iot-in-the-5g-era. (accessed on 16 September 2021).
  14. Ushakov, Y.; Ushakova, M.; Legashev, L. Problems of Building Infrastructure Vehicular Ad Hoc Networks Based on SD-WAN Technologies. In Proceedings of the 2022 International Siberian Conference on Control and Communications (SIBCON), 17–19 November 2022; pp. 1–4. [Google Scholar]
  15. 5G Americas. The 5G Evolution: 3GPP Releases 16–17. White Paper. 2020. Available online: https://www.5gamericas.org/5g-evolution-3gpp-releases-16-17/ (accessed on 16 September 2021).
  16. Datenfluss. 5G RAN CU–DU Network Architecture, Transport Options and Dimensioning. NGMN, 12 April 2019. Available online: https://www.ngmn.org/publications/5g-ran-cu-du-network-architecture-transport-options-and-dimensioning.html (accessed on 16 September 2021).
  17. Boutaba, R.; Salahuddin, M.A.; Limam, N.; Ayoubi, S.; Shahriar, N.; Estrada-Solano, F.; Caicedo, O.M. A Comprehensive Survey on Machine Learning for Networking: Evolution, Applications, and Research Opportunities. J. Internet Serv. Appl. 2018, 9, 16. [Google Scholar] [CrossRef]
  18. Tayyaba, S.K.; Khattak, H.A.; Almogren, A.; Shah, M.A.; Din, I.U.; Alkhalifa, I.; Guizani, M. 5G Vehicular Network Resource Management for Improving Radio Access Through Machine Learning. IEEE Access 2020, 8, 6792–6800. [Google Scholar] [CrossRef]
  19. Hou, L.; Lei, L.; Zheng, K.; Wang, X. A Machine-Learning-Based Proactive Caching Strategy for Non-Safety Related Services in Vehicular Networks. IEEE Internet Things J. 2019, 6, 4512–4520. [Google Scholar] [CrossRef]
  20. Tuxera. Autonomous Cars Generate More than 300 TB of Data per Year, 2021. Available online: https://www.tuxera.com/blog/autonomous-cars-300-tb-of-data-per-year/ (accessed on 21 February 2024).
  21. Richards, M. Software Architecture Pattern: Understanding Common Architecture Patterns and When to Use Them; O’Reilly: Sebastopol, CA, USA, 2015. [Google Scholar]
  22. Yang, P.; Xi, X.; Quek, T.Q.; Chen, J.; Cao, X.; Wu, D. RAN Slicing for Massive IoT and Bursty URLLC Service Multiplexing: Analysis and Optimisation. IEEE Internet Things J. 2021, 8, 14258–14275. [Google Scholar] [CrossRef]
  23. Du, P.; Nakao, A. Deep Learning-Based Application Specific RAN Slicing for Mobile Networks. In Proceedings of the 2018 IEEE 7th International Conference on Cloud Networking (CloudNet), Tokyo, Japan, 9–11 October 2018; pp. 1–3. [Google Scholar] [CrossRef]
  24. Kazmi, S.H.A.; Qamar, F.; Hassan, R.; Nisar, K. Routing-Based Interference Mitigation in SDN-Enabled Beyond 5G Communication Networks: A Comprehensive Survey. IEEE Access 2023, 11, 4023–4041. [Google Scholar] [CrossRef]
Figure 1. Proposed architecture. Coloured dots on vehicles represent critical IoT sensors (e.g., Camera, LiDAR, and RADAR) commonly embedded in connected vehicles.
Figure 1. Proposed architecture. Coloured dots on vehicles represent critical IoT sensors (e.g., Camera, LiDAR, and RADAR) commonly embedded in connected vehicles.
Network 05 00028 g001
Figure 2. SDN-Enabled Vehicular Edge Network. Differently coloured sensor nodes represent service differentiation for critical IoT sensors with varying QoS requirements. Solid arrows illustrate the possible direction of communication flow on the network, whilst dotted arrows indicate local real-time data processing.
Figure 2. SDN-Enabled Vehicular Edge Network. Differently coloured sensor nodes represent service differentiation for critical IoT sensors with varying QoS requirements. Solid arrows illustrate the possible direction of communication flow on the network, whilst dotted arrows indicate local real-time data processing.
Network 05 00028 g002
Figure 3. Event-Driven SD-WAN framework for identity, data and dynamic access management in 5G communication systems. Solid arrows illustrate the direction of communication flow on the network, whilst dotted arrows indicate real-time communication between the control and data plane.
Figure 3. Event-Driven SD-WAN framework for identity, data and dynamic access management in 5G communication systems. Solid arrows illustrate the direction of communication flow on the network, whilst dotted arrows indicate real-time communication between the control and data plane.
Network 05 00028 g003
Figure 4. System flow diagram.
Figure 4. System flow diagram.
Network 05 00028 g004
Figure 5. Average packets transmitted by a vehicle to the gNB.
Figure 5. Average packets transmitted by a vehicle to the gNB.
Network 05 00028 g005
Figure 6. Packet transmission vs. reception.
Figure 6. Packet transmission vs. reception.
Network 05 00028 g006
Figure 7. Average flow duration.
Figure 7. Average flow duration.
Network 05 00028 g007
Figure 8. Average flow rate at SDN gateway.
Figure 8. Average flow rate at SDN gateway.
Network 05 00028 g008
Figure 9. Average flow rate at gNB.
Figure 9. Average flow rate at gNB.
Network 05 00028 g009
Figure 10. Average delay at the SDN gateway.
Figure 10. Average delay at the SDN gateway.
Network 05 00028 g010
Figure 11. Average delay at gNB.
Figure 11. Average delay at gNB.
Network 05 00028 g011
Figure 12. Event-driven multi-queue and delay broker framework. Queues represent virtualised network components (e.g., SDN gateway, gNB, edge servers) that form the distributed 5G Infrastructure. Resource groups denote the corresponding flow rules and tables deployed for data processing.
Figure 12. Event-driven multi-queue and delay broker framework. Queues represent virtualised network components (e.g., SDN gateway, gNB, edge servers) that form the distributed 5G Infrastructure. Resource groups denote the corresponding flow rules and tables deployed for data processing.
Network 05 00028 g012
Figure 13. Packet delivery rate comparison with existing SDN-based approach in [14].
Figure 13. Packet delivery rate comparison with existing SDN-based approach in [14].
Network 05 00028 g013
Figure 14. Traffic classification comparison with existing AI-based methods in [18,23].
Figure 14. Traffic classification comparison with existing AI-based methods in [18,23].
Network 05 00028 g014
Table 1. Arrival and service rates at the SDN gateway and VXLAN link to gNB (PPS).
Table 1. Arrival and service rates at the SDN gateway and VXLAN link to gNB (PPS).
Sensor TypeTEST 1 (PPS)TEST 2 (PPS)TEST 3 (PPS)
CAMERA2.34.66.6
LIDAR1.42.84.0
RADAR0.61.31.9
Table 2. Service rates at gNB (Packets per Second (PPS)).
Table 2. Service rates at gNB (Packets per Second (PPS)).
Sensor TypeTEST 1 (PPS)TEST 2 (PPS)TEST 3 (PPS)
CAMERA8.414.718
LIDAR4.39.011
RADAR2.44.45.9
Table 3. System utilisation at gNB ( ρ ).
Table 3. System utilisation at gNB ( ρ ).
Sensor TypeTEST 1 ( ρ )TEST 2 ( ρ )TEST 3 ( ρ )
CAMERA0.270.310.36
LIDAR0.330.310.36
RADAR0.250.300.32
Table 4. Average packets transmitted by a single vehicle (packets).
Table 4. Average packets transmitted by a single vehicle (packets).
Sensor TypeTEST 1 (Packets)TEST 2 (Packets)TEST 3 (Packets)
CAMERA194519481940
LIDAR116711701169
RADAR778780780
Table 5. Average delay for all transmitted packets by the SDN gateway (s).
Table 5. Average delay for all transmitted packets by the SDN gateway (s).
Sensor TypeTEST 1 (S)TEST 2 (S)TEST 3 (S)
CAMERA845.7423.5294
LIDAR833.6417.9292.3
RADAR1296.7600410.5
Table 6. Average delay per packet transmitted at SDN gateway (s).
Table 6. Average delay per packet transmitted at SDN gateway (s).
Sensor TypeTEST 1 (S)TEST 2 (S)TEST 3 (S)
CAMERA0.430.220.15
LIDAR0.710.360.25
RADAR1.670.770.50
Table 7. Average packets offloaded for all four vehicles (packets).
Table 7. Average packets offloaded for all four vehicles (packets).
Sensor TypeTEST 1 (Packets)TEST 2 (Packets)TEST 3 (Packets)
CAMERA778077927760
LIDAR466846804676
RADAR311231203120
Table 8. Average delay at gNB during offloading (s).
Table 8. Average delay at gNB during offloading (s).
Sensor TypeTEST 1 (S)TEST 2 (S)TEST 3 (S)
CAMERA931.8530.8430.6
LIDAR1293.3520.7420.6
RADAR1092.5711.4530.9
Table 9. Average delay per packet at gNB during offloading (s).
Table 9. Average delay per packet at gNB during offloading (s).
Sensor TypeTEST 1 (S)TEST 2 (S)TEST 3 (S)
CAMERA0.120.070.06
LIDAR0.280.110.09
RADAR0.350.230.17
Table 10. Overall delay per packet for vehicular data pre-processing and gNB offloading (s).
Table 10. Overall delay per packet for vehicular data pre-processing and gNB offloading (s).
Sensor TypeTEST 1 (S)TEST 2 (S)TEST 3 (S)
CAMERA0.550.290.21
LIDAR0.990.470.34
RADAR2.0210.67
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Ifaloye, A.; Takruri, H.; Al-Zaidi, R. Towards Intelligent 5G Infrastructures: Performance Evaluation of a Novel SDN-Enabled VANET Framework. Network 2025, 5, 28. https://doi.org/10.3390/network5030028

AMA Style

Ifaloye A, Takruri H, Al-Zaidi R. Towards Intelligent 5G Infrastructures: Performance Evaluation of a Novel SDN-Enabled VANET Framework. Network. 2025; 5(3):28. https://doi.org/10.3390/network5030028

Chicago/Turabian Style

Ifaloye, Abiola, Haifa Takruri, and Rabab Al-Zaidi. 2025. "Towards Intelligent 5G Infrastructures: Performance Evaluation of a Novel SDN-Enabled VANET Framework" Network 5, no. 3: 28. https://doi.org/10.3390/network5030028

APA Style

Ifaloye, A., Takruri, H., & Al-Zaidi, R. (2025). Towards Intelligent 5G Infrastructures: Performance Evaluation of a Novel SDN-Enabled VANET Framework. Network, 5(3), 28. https://doi.org/10.3390/network5030028

Article Metrics

Back to TopTop