Performance Analysis of Internet of Vehicles Mesh Networks Based on Actual Switch Models

Abstract

The rapid growth of the automotive industry has exacerbated the conflict between the complex traffic environment, increasing communication demands, and limited resources. Given the imperative to mitigate traffic and network congestion, analyzing the performance of Internet of Vehicles (IoV) mesh networks is of great practical significance. Most studies focus solely on individual performance metrics and influencing factors, and the adopted simulation tools, such as OPNET, cannot achieve the dynamic link generation of IoV mesh networks. To address these problems, a network performance analysis model based on actual switches is proposed. First, a typical IoV mesh network architecture is constructed and abstracted into a mathematical model that describes how the link and topology changes over time. Then, the task generation model and the task forwarding model based on actual switches are proposed to obtain the real traffic distribution of the network. Finally, a scientific network performance indicator system is constructed. Simulation results demonstrate that, with rising task traffic and decreasing node caching capacity, the packet loss rate increases, and the task arrival rate decreases in the network. The proposed model can effectively evaluate the network performance across various traffic states and provide valuable insights for network construction and enhancement.

1. Introduction

In recent years, with the development of intelligent devices and wireless communication technology, the communication distance of devices has gradually increased and has steadily expanded, encompassing diverse geographical terrains and different types of communication services. Numerous communication devices are interconnected, and play an important role in various fields. With the increase in the global urban population and the rapid development of transportation infrastructures, the number of vehicles has grown rapidly and become an important means of transport for people to travel. In this context, recurrent traffic congestion and accidents, traffic inefficiency, and other problems are becoming increasingly serious. Road safety is now a significant societal issue. Intelligent Traffic System (ITS) establishes a holistic intelligent integrated transportation system by integrating advanced technologies, and the application of this system is an effective strategy to solve the above problems. Among them, the IoV mesh network, as an important part of ITS, carries substantial responsibility for information exchange and is the key to the operation of the whole system [1].

In contrast to conventional mobile communication networks, the IoV mesh network utilizes vehicles with interconnected interface devices as network nodes. According to the different communication objects, the communication links in vehicular networks include Vehicle-to-Vehicle (V2V) communication and Vehicle-to-Infrastructure (V2I) communication [2]. Among them, V2V communication facilitates the process of information exchange among vehicles within their communication range. V2I communication realizes the information interaction between vehicles and the Road Side Unit (RSU). V2I communication enables RSUs to offer services like collision warnings, blind spot detection, dynamic route planning, and other vehicle–road cooperative services. Vehicles in IoV mesh networks have sensors and other devices that can autonomously collect information and transmit it via V2V communication links to enable data interaction between vehicles. Therefore, each node of the vehicle cluster follows the rules of autonomy; cluster members cooperate using information transmission and sharing, showing strong self-organizational capabilities, and can handle tasks collaboratively [3]. Based on the above analysis, unlike individual vehicles, IoV mesh networks are richer in resources and can handle more complex tasks and provide better services. However, at the same time, environmental interference, vehicle density and distribution, and high-speed mobility also pose challenges for scheduling, optimization, and decision-making of IoV mesh networks [4]. Addressing the bottleneck of collaborative communication modeling in IoV mesh networks is a current research focus in IoV mesh network construction.

To optimize the problem of IoV mesh network construction, it is firstly necessary to comprehensively analyze the various performances of IoV mesh networks and comprehend how various network parameters affect performance. Among the factors affecting the performance of IoV mesh networks, the primary factor is task traffic [5]. Because of the high-speed mobility and uneven distribution of vehicles, the topology in IoV mesh networks rapidly changing, leading to uncertain task forwarding and transmission. With exceptionally high vehicle density, the task traffic in the network rises sharply and could potentially result in node cache congestion. Therefore, it is of great practical significance to study the task model and performance metrics system in IoV mesh networks and explore the traffic distribution.

In addressing vehicular network performance, many researchers have carried out studies on system analysis modeling. For example, Killat et al. [6] assessed performance under the 802.11p standard using four factors: communication distance, node transmission power, packet transmission rate, and task flow, as input variables in order to examine the packet reception rate. However, the stationary nodes in the network are not suitable for the high-speed mobility characteristic of IoV mesh networks. Huang et al. [7] proposed the Real Vehicular Wireless Network model, which allowed for a more realistic capacity analysis in vehicular networks. The actual geometry of the urban area was represented by a Euclidean planar graph, so as to analyze the interference relations and calculate the asymptotic capacity. Kwon et al. [8] represented the network as a geometric structure of straight lines and points for a one-dimensional V2V network with topology changing over time and analyzed network connectivity and capacity using geometric probability. Through mathematical modeling, it was found that the network capacity increases in a specific way, which verifies the validity of the analytical results. Chen et al. [9] analyzed the reachable throughput of a vehicular network with finite traffic density under a cooperative communication strategy. By obtaining a closed-form expression for the reachable throughput, they unveiled the relationship between throughput and key performance-influencing parameters like communication transmission rate and vehicle density. The above study obtained the expressions for solving the network throughput through mathematical modeling, but did not account for queuing, caching, and task-forwarding processes at nodes in actual networks.

Zhang et al. [10] built a vehicle movement model to investigate the network coverage problem to ensure stable overall performance. They also compared the performance of network throughput, average network delay, packet loss rate, average routing hops, etc. under different protocols. Saadallah et al. [11] conducted a performance analysis of connectivity and throughput of vehicular networks on motorways, exploring its correlation to vehicle density, number of lanes, etc. The results, backed up by simulations of realistic vehicular traffic, showed that the V2V + V2I architecture can significantly improve connectivity. Aljabry et al. [12] investigated the performance of a common reactive routing protocol where two scenarios are considered. The first scenario was a comparison between V2V and V2I communication modes. The second one was made between two maps, Basrah city and the Manhattan grid, in V2V mode. The implementation of the routing protocol revealed a comparative results analysis using Quality of Service (QoS) parameters, such as network throughput. However, these studies were conducted using prevalent network simulation tools (e.g., OPNET and NS), which cannot accurately address the dynamic link generation problem and portray time-varying features, and they do not have a suitable methodology to efficiently calculate key performance metrics such as node load rate. Within IoV mesh networks, link construction, disruption, and recovery among nodes constantly change, and the performance of individual nodes is also closely related to the network performance. In addition, based on the actual switches, packet forwarding relies on the port routing forwarding table at the nodes and undergoes a series of complex processes. Therefore, in order to solve the above problems, this paper proposes a task-forwarding strategy based on the actual switch model, on the basis of which a network indicator system is constructed to analyze the performance of IoV mesh networks under different vehicle densities, vehicle caching capacities, etc., which lays the foundation for load-balanced routing design, task offloading, and resource allocation in IoV mesh networks. The specific research is as follows:

(1) We construct a typical network architecture for IoV mesh networks and abstract it into a mathematical model. We create V2V and V2I channel models and vehicle movement models for dynamic link generation and interruption to obtain the network topology over time.

(2) The task generation and forwarding model are proposed to obtain the actual traffic distribution of the network, through user traffic modeling and the construction of node port routing tables to establish task scheduling, forwarding, and transmission strategies based on the actual switch model.

(3) Based on the traffic transmission results, a network performance indicator system is established. We conduct several experimental simulations to assess variations in multiple indicators across diverse vehicle densities and caching capacities, thereby validating the validity and accuracy of the proposed model.

2. Related Works

Since the pioneering work of Gupta and Kumar in 2000 [13], an increasing amount of research has focused on the performance of wireless communication networks. As elements such as the spatial distribution of network nodes, wireless channel fading characteristics, and other factors influence the interference distribution in wireless networks, current analyses of wireless multi-hop networks primarily concentrate on interference and outage probability, as well as capacity and throughput assessments. For example, Vinh et al. [14] derived the system outage probability based on a specific fading model for a relayed two-hop collaborative relaying system containing multiple relay nodes for a pair of communication users using amplify-and-forward and decode-and-forward protocols, respectively. Olmedo et al. [15] proposed a wireless TCP protocol improvement considering Negative Acknowledgment (NACK), which allowed for distinguishing between losses due to congestion and losses due to wireless channel problems. This method improved QoS and throughput. Li et al. [16] analyzed the outage probability and throughput of a full-duplex orthogonal frequency-division-multiplexing relay network based on the amplify-and-forward protocol and calculated the energy harvesting duration and the signal transmission time, and the capacity and throughput. The optimal time slot interval between energy harvesting duration and signal transmission time was used as a way to maximize the network throughput.

However, the research on wireless communication network performance has generally overlooked the influence of mobility on network layout and node distribution. Consequently, analyzing the performance of IoV mesh networks requires an exploration of the relationship between system parameters and network performance in the context of system characteristics. Rahmani et al. [17] conducted a comparison of various network topologies for designing a suitable IP/Ethernet network for audio and video communication in vehicles, considering QoS performance and production costs. Different vehicular networks’ capacities were analyzed in the literature [18,19,20], with the capacity determined using the proportionality criterion approach. Distinctions exist among these studies; Nekoui et al. [18] assumed all vehicles travel along a single road, while Lu et al. [19] assumed that there are multiple roads in the network, but the vehicles can only move around the designated centers. Wang et al. [20] expanded on these scenarios by allowing vehicles to move freely within the network area. By maximizing the throughput of information transfer between vehicles and RSUs, efficient information transfer path algorithms were derived from this model. Sarvade et al. [21] compared the latest IEEE standard 802.11ac with previous MAC protocols based on parameters such as throughput, jitter, and end-to-end delay using real traffic pattern scenarios. Realistic traffic patterns were generated using SUMO, and NS-3 evaluated protocol performance. Additionally, Lai et al. [22] examined the average packet loss rate’s impact under distributed relay selection for multihop broadcasting in vehicular networks, taking into account vehicle mobility, wireless channel conditions, and media access control. They proposed an analytical model to analyze the average packet loss rate.

Zhao et al. [23] introduced a model for analyzing the QoS and capacity of vehicular networks on one-dimensional motorways and two-dimensional intersection roads. The validation of the proposed model was conducted through NS-2 simulations and further expanded to include the derivation of additional QoS metrics like packet reception probability, packet acceptance rate, and broadcast link capacity. Han et al. [24] proposed a distance-weighted back-pressure dynamic routing (DBDR). Simulations in a practical road scenario using NS-2 and VanetMobiSim showed that DBDR outperforms the existing protocols in terms of packet delivery ratio, throughput, and average packet delay when the network becomes congested. Jiang et al. [25] constructed a new software-defined networking-based IoV heterogeneous networking measurement framework and proposed a performance measurement and analysis method. The performance indexes and measurement methods for the loss, throughput, etc. were derived in detail. The switch selection mechanism and packet sampling process were used to establish optimal measurement points of advantage and quickly obtain the needed measurement information. Wang et al. [26] introduced a tree-cut mapping-based average maximum flow solution method (TCMANF) and utilized average network flow (ANF) as a metric to assess the overall network quality of a city. TCMANF quickly determined the maximum flow rate within the inter-node network to obtain the average network flow rate. Zheng et al. [27] combined V2V and V2I communication modes to propose a framework for assessing the communication capabilities of vehicular networks operating in mixed traffic scenarios. They introduced a predictive communication strategy to enhance vehicular network capacity in mixed-traffic by pre-caching necessary content in the infrastructure based on predicted vehicle trajectories. Gupta et al. [28] used IEEE 802.11p and IEEE 802.11s for static and moving vehicles working under mesh topology. The performance evaluation was accomplished by simulation on NS-3. Throughput, packet delivery functions and packet sizes are parameters that have been evaluated. Malnar et al. [29] developed an NS-3-based vehicular network framework to analyze performance metrics such as throughput and packet loss. Several topology-based routing protocols were compared and Expected Transmission Number (ETX) metric was proposed to improve performance. Park et al. [30] designed the Domain-based In-vehicle network Architecture (DIA) and the Zone-based In-vehicle network Architecture (ZIA) for autonomous driving, and then analyzed and validated the superiority of the ZIA and its suitability as an automotive architecture using the OMNeT++ network simulator.

The above research works investigated the performance of vehicular networks in terms of deriving theoretical expressions and modeling traffic on simulation tools. However, most of them only considered a single metric, and multiple metrics were mostly studied for network capacity and packet loss rate. Due to the high-speed mobility of vehicles in IoV mesh networks and the diversity of user tasks, the topological connectivity relationships are changing rapidly, and performance analysis needs to be considered from multiple metrics. In addition, the performance of IoV mesh networks is affected not only by vehicle density, which is a factor from the overall perspective of the network, but also by the capacity of the nodes, especially the caching capacity. Furthermore, common simulation tools such as OPNET and NS cannot accurately solve the dynamic generation of links, i.e., the time-varying characteristics of links. Tools such as NS-3 do not support metrics other than hop count, and there is no suitable method to efficiently calculate key performance indicators, such as node load rate [29]. Therefore, based on the above problems, this paper constructs the task-forwarding model based on actual switches to analyze the network performance in terms of multiple metrics and multiple factors.

3. Network Architecture and System Model

The IoV mesh network constitutes a vital component of ITS. Within this network, each vehicle is equipped with communication and computing devices. Vehicles possess the capability to directly communicate with each other while in motion at high velocities, even in situations where communication infrastructure such as base stations and wireless access points are unavailable nearby. It is essential to recognize that, within an IoV mesh network, vehicles not only transmit and receive data, but also serve as routers. Consequently, when the sender node is distant from the receiver node, the sender node transmits data through neighboring nodes in a multi-hop manner. The IoV mesh network architecture aims to facilitate communication among proximate vehicles, as well as between vehicles and adjacent facilities, which can be divided into two parts: inter-vehicle communication (V2V) and communication between vehicles and infrastructure (V2I). A typical IoV mesh network architecture is shown in Figure 1.

We consider a simplified motion scenario where the newly added vehicles are not involved in the task and, hence, the total number of network nodes remains unchanged for the simulation duration. We set the starting position of the originally existing vehicle that will not leave the simulation area based on its maximum movement speed.

Each vehicle and RSU autonomously generates diverse tasks. For example, when a vehicle encounters factors such as occlusion or bad weather, and is unable to make a correct judgment on the current traffic light or the traffic light changes in the coming period, the RSU can send traffic light information to the neighboring vehicles, and the vehicle can clarify the traffic light status and forward it to other vehicles after obtaining this information. At the same time, it can also determine the time for the vehicle to arrive at the intersection according to its own location and the map, so as to realize traffic light speed guidance. Vehicle tasks include augmented reality, real-time video analytics, and human behavior recognition for drivers, pedestrians, etc.

When a vehicle or RSU requires task collection and transmission, it sends requests to neighboring vehicles or RSUs so that different tasks have different destinations. This framework enables vehicles and RSUs to engage in communication, information exchange, collaborative task assignment, and data processing. Such collaboration allows for real-time environmental sensing and timely adaptation of collaboration strategies across diverse application scenarios. If the network is faulty (e.g., node failure or link outage), there may be an impact on network performance. For example, when a task has been transmitted to a node, but that node suddenly fails and is unable to process and forward the task, at this point, the packet loss rate of the network will increase, the task arrival rate will decrease, etc.

3.1. Network Model

In order to analyze network performance, this subsection constructs the IoV mesh network dynamic topology. Since the topology depends on link construction, the IoV mesh network channel model is initially established by abstracting the network architecture into a mathematical model. The scenario involves multiple vehicles and RSUs distributed at an intersection with bi-directional traffic flow. To determine the boundaries of the network model, we refer to the urban case described in 3GPP TR 36.885 [31]. The size of the four grids included is 866 m × 500 m, which has four lanes (two lanes in both forward and reverse directions), and the lane width is 3.5 m. The scenario model diagram is shown in Figure 2.

To facilitate the modeling, a spatial right-angled coordinate system is established with the center of the road as the coordinate origin, the xoy plane parallel to the ground, and the z-axis perpendicular to the ground. $U=\{u_1,\cdots ,u_p\}$ is defined as the set of vehicles and $S=\{s_1,\cdots ,s_q\}$ represents the set of RSUs, where p is the total number of vehicles and q is the total number of RSUs. The IoV mesh network model is shown in Figure 3.

$(x_i,y_i,z_i)$ and $(x_j,y_j,z_j)$ are the coordinates of any two nodes i and j in the network, respectively. Therefore, the Euclidean distance $d_{ij}$ between any two nodes at moment t can be given by:

$$d_{ij}\left(t\right)=\sqrt{{(x_i\left(t\right)-x_j\left(t\right))}^2+{(y_i\left(t\right)-y_j\left(t\right))}^2+{(z_i\left(t\right)-z_j\left(t\right))}^2}$$

(1)

According to the destination, the network has two ways of transmitting data, which are V2I and V2V. When V2V communication is performed, based on existing communication protocols, each vehicle can exchange information with other vehicles located within its maximum access range denoted by D. From this, the condition that defines the existence of a link between vehicles is shown below, where $g_{ij}\left(t\right)=1$ indicates that the link is connected and, vice versa, interrupted.

$$g_{ij}^{v2v}\left(t\right)=\left\{\begin{array}{cc}1,\hfill & d_{ij}^{v2v}(t)\le D_j\hfill \\ 0,\hfill & otherwise\hfill \end{array}\right.$$

(2)

Therefore, the link capacity at moment t is shown below, where ${B_{ij}}^{v2v}$ and ${h_{ij}}^{v2v}$ are the channel bandwidth and the channel gain between vehicle i and vehicle j, respectively, and ${\sigma }^2$ denotes the background noise power. Moreover, ${P_i}^{v2v}$ denotes the transmit power of vehicle i, and $A_0$ = −17.8 dB is a constant parameter.

$$R_{ij}^{v2v}(t)=g_{ij}^{v2v}(t)·B_{ij}^{v2v}{log}_2(1+\frac{P_i^{v2v}·h_{ij}^{v2v}}{{\sigma }^2+A_0{(d_{ij}^{v2v}(t))}^{-2}})$$

(3)

Similarly, when V2I communication is performed, each RSU can exchange information with vehicles located within its maximum access range. Assuming that the network has been pre-allocated with orthogonal spectrum resources and there is no interference, at this time, the link capacity at moment t of V2I is shown in Equation (4). $B_{ij}^{v2I}$ and $h_{ij}^{v2I}$ are the channel bandwidth and the channel gain between vehicle i and RSU j, respectively [32].

$$R_{ij}^{v2I}(t)=g_{ij}^{v2I}(t)·B_{ij}^{v2I}{log}_2(1+\frac{P_i^{v2I}·h_{ij}^{v2I}}{{\sigma }^2})$$

(4)

Since this paper focuses on network traffic analysis, assuming all vehicles move in a uniform linear motion for modeling simplicity. A given time interval T is divided into n time slots, and the length of the time slot is $\Delta t=T/n$. The network topology remains unchanged within each time slot. However, as vehicles move, connections between vehicles in different time slots change. The weighted adjacency matrix of the network of the k th time slot is defined by Equation (5), where ${\pi }_{ij}^k$ represents the link rate. ${\pi }_{ij}^k=0$ indicates no connectivity. In addition, due to changes in the link state, data may be cached on a node during transmission, waiting for link connection, i.e., the data caching process.

$$G_k=\left[\begin{array}{cccc}0& {\pi }_{12}^k& \cdots & {\pi }_{1p}^k\\ {\pi }_{21}^k& 0& \cdots & {\pi }_{2p}^k\\ ⋮& ⋮& \ddots & ⋮\\ {\pi }_{p1}^k& {\pi }_{p2}^k& \cdots & 0\end{array}\right]$$

(5)

Consequently, the dynamic topology of the IoV mesh network at each moment can be obtained through the above analyses, serving as a foundation for task traffic transmission and performance analysis.

3.2. Task Generation Model

Since the main service object of the IoV mesh network is vehicle users, and the network performance will affect the service experience of users, it is significant to analyze quantitatively the quality of user service under different network conditions. For this reason, this subsection proposes a method for modeling and analyzing traffic with a focus on user experience. First, the traffic model is constructed, and the traffic distribution model based on actual switch port forwarding is investigated for analyzing the real-time distribution of data flow within the network, which forms the basis for the subsequent construction of the indicator system and performance analysis.

Before modeling the traffic generation of vehicles or RSUs, this subsection constructs the port routing table of each node based on the forwarding strategy of the actual switch, which provides the basis for the generation of task routes. By traversing the network nodes and utilizing Dijkstra’s shortest path algorithm, a route originating from one node to another network node is created; this route is then converted into a port route to facilitate task forwarding by consulting the node’s routing table. This is shown in Figure 4. In the topology on the left side of the picture, the values on the links denote the path distance so that the route for Task 1 can be obtained by the Dijkstra algorithm as 1→2→4→6. In the topology on the right side of Figure 3, ‘1 2’ denotes the port utilized by node 1 to communicate with node 2. The number of ports usually is finite.

Then, the traffic generation of vehicles or RSUs is modeled to obtain the task distribution. The procedure is outlined as follows: (1) Divide the IoV mesh network into multiple grid spaces and count the number of vehicles and RSUs initiating tasks within each grid. (2) Randomly select the destination of each initiating task node. (3) Construct a traffic model to generate task routes for each initiating task node at each moment, ensuring task QoS values meet the criterion of the average packet loss rate across the network of less than 5%. (4) The network exists the high, medium and low priority tasks.

3.3. Task Forwarding Model

To address the traffic forwarding issue in IoV mesh networks and gain insights into traffic distribution within the network, this subsection proposes a forwarding model based on actual switches to analyze the network traffic in different states.

Initially, the analysis is performed for a single service. Within an IoV mesh network with p nodes, the switch model for forwarding a single task at each vehicle node is shown in Figure 5. Here, $\lambda$ represents the input traffic of the task at the node, $\mu$ denotes the output traffic (i.e., the forwarded traffic through the switch), C indicates the node’s cache capability. The cache refers to the data exchange buffer on the network layer of the node switch, also called the packet buffer size, which is a queuing structure that is used by the switch to coordinate speed-matching problems between different network devices. When a switch does store-and-forward, it presses the packets in the buffer into the egress queue for transmission. L denotes the actual cache queue length of the node. Furthermore, the node’s performance incorporates its forwarding capacity, represented by F. When the switch forwards, tasks in the cache queue are taken out and pressed into the egress queues of different ports according to rules, and the egress queue length is the egress link rate.

Based on Section 3.2, the per-hop port routing of the task can be obtained. Simultaneously, combined with the network topology, we can ascertain the link rate denoted as R between the current location node of the task and the subsequent hop node. Consequently, two distinct scenarios emerge for forwarding traffic of a single task:

(1) $\lambda +L\le C$: At this point, the sum of the actual cache queue length of the task node and the task input traffic is less than or equal to the node’s cache capacity. Therefore, the task has no packet loss on this node, and the output traffic is constrained by the node’s forwarding capacity and the forwarding link rate, i.e., $\mu =min\{\lambda +L,F,R\}$. The actual cache queue length is further updated as $L^{\prime }=\lambda +L-\mu$.

(2) $\lambda +L>C$: At this point, the sum of the actual cache queue length of the task node and the task input traffic is greater than the node’s cache capacity. So the task will lose some packets at that node, and the number of packets lost is represented by $Loss=\lambda +L-C$. In this case, the task enters the switch and the packet loss occurs first. Furthermore, the node’s cache is full, and thus the switch’s forwarded traffic is $\mu =min\{C,F,R\}$, and the actual cache queue length is further updated to be $L^{\prime }=\lambda +L-Loss-\mu$.

Once $\mu$ has been obtained, it is necessary to calculate the input traffic at the next hop node. The input traffic is influenced by the actual link rate denoted as $R^{\prime }$, as switches are incapable of sensing interruptions or reductions in link rate (i.e., changes in egress queue lengths) caused by environmental fluctuations. In conjunction with the changing real network topology, the following two scenarios exist:

(1) $\mu \le R^{\prime }$: At this point, $\mu$ is less than or equal to the number of packets that the link can actually transmit per unit of time, so the number of packets lost on the link is 0, and the input traffic to the next hop node is $\lambda =\mu$.

(2) $\mu >R^{\prime }$: At this point, $\mu$ is greater than the number of packets that the link can actually transmit per unit of time, and packet loss will occur on the link, with the number of packets lost being $Los{s}^{\prime }=\mu -R^{\prime }$. Therefore, the input traffic to the next hop node is $\lambda =R^{\prime }$.

For the multi-task scenario, when scaling single-task traffic to multi-task traffic, traffic on different ports on the node will be overlapped. An example of three nodes with two tasks is as follows.

As shown in Figure 6 and Figure 7, at a specific moment, node 1 has two tasks at the same time, whose F and C are 300 packet/s and 200 packet, respectively. Task 1 has 100 cached packets on the node, and task 2 is a newly input task with 200 packets. According to the node port routing table and task destinations, it is assumed that the next-hop routes for the two tasks are node 2 and node 3.

At this point, node 1 initiates the sequential arrangement of the two tasks. Tasks are ordered based on a prioritization scheme where high-priority tasks take precedence, followed by medium-priority tasks, then low-priority tasks, while tasks of the same priority are randomized. Then, node 1 assigns forwarding and caching capacities to tasks. F and C are allocated based on the sorting result; the more advanced task in the queue is preferred to be allocated with the same number of F and C as its packets until the allocation is completed. Therefore, the forwarding and caching sizes of task 1 are 100 and 100, respectively, and the sizes of task 2 are 200 and 100, respectively. In cases where multiple tasks are present in the egress link simultaneously, each task is assigned the transmission rate R according to the same prioritization principle employed for F. Upon completion of the assignment, the output flow of each task can be compared and calculated according to the single-task forwarding model, so as to realize the traffic forwarding of multiple tasks.

Based on the above analysis, when multiple nodes in the network handle multiple tasks, the methodology outlined can be employed to compute the input flow, output flow, cache flow, etc., for each node. This involves directing the traffic from each specific task to its designated forwarding port, allowing for seamless forwarding in conjunction with routing. It ultimately achieves the traffic simulation of the overall network, so as to explore the distribution of the network traffic in different conditions. This approach establishes a groundwork for further investigations into network performance metrics, such as packet loss rates.

4. Network Indicator System Construction

Due to the intricate physical and logical interconnections across the various levels of the IoV network, consequently, to effectively analyze the performance of the network, comprehend the task propagation patterns, and further clarify the relationship between the various levels of the system, it is imperative to establish a comprehensive assessment indication framework. This framework will facilitate the analysis of the IoV network’s performance, check the level of congestion in a highly dynamic network, and enhance the scientific and anticipatory management of IoV networks.

Based on Section 3, after completing the construction of the network traffic transmission model, the results of the forwarding algorithm can be obtained, and massive output data can be processed to generate multiple core indicators for the construction of the evaluation system. Given that IoV networks primarily serve users, the network performance significantly impacts their service experience, which varies according to users’ priority. Therefore, from the task perspective, the successful task arrival rate, packet loss rate and delay are selected as the core indicators; from the network perspective, when the task is transmitted in the network, the load rate of the nodes and links are different, and both of them will indirectly affect user task service experiences. For example, when the vehicle initiating the service overloads due to failure or excessive link loads between it and the next-hop node, it leads to task accumulation and packet loss. Therefore, using the node load rate, link load rate, and the actual total task traffic of the network as the core indicators can effectively reflect the blocking situation of the network for performance evaluation.

In summary, considering that the analysis indicators should be scientific, reasonable and complete, from the perspective of network structure and performance, this paper adopts five indicators to analyze the network performance. These indices include packet loss rate, task arrival rate, link load rate, node load rate, and total network traffic.

4.1. Packet Loss Rate

The packet loss rate is a critical metric for assessing the reliability of network transmission, and serves as an essential indicator of network performance. High packet loss rates indicate network abnormalities. By tracking the number of packets dropped by each task at every vehicle node, the packet loss rate of each task can be calculated, leading to the overall packet loss rate for the network. Specifically, assuming that the number of lost packets and the total packets of network tasks are $\{los{s}_1,\cdots ,los{s}_n\}$, $\{{\phi }_1,\cdots ,{\phi }_n\}$, respectively, where n is the total number of tasks at the current moment, the overall packet loss rate of the network is:

$$loss\_rate=\frac{\frac{los{s}_1}{{\phi }_1}+\cdots +\frac{los{s}_n}{{\phi }_n}}{n}$$

(6)

4.2. Task Arrival Rate

A series of network issues during data transmission can result in some tasks losing packets or accumulating at nodes, thereby decreasing the success rate of tasks reaching their intended destinations. Therefore, the successful task arrival rate is a crucial parameter in assessing network performance, as it reflects the reliability of data transmission. This indicator is calculated as the ratio of packets successfully reaching the destination to the total number of packets sent at the initiation of the task. The formula is as follows:

$$arrive\_rate=\frac{\sum \frac{{\gamma }_i}{{\rho }_i}}{M}$$

(7)

where ${\gamma }_i$ denotes the total number of packets at which the task i reaches the destination, ${\rho }_i$ denotes the total number of packets at which the task i is initiated, and M represents the total number of generated tasks. A lower task arrival rate indicates that most of tasks are congested in the network or packet loss occurs, which will significantly affect the transmission performance of the IoV network and even cause complete network failure.

4.3. Node Load Rate

The node load rate is defined as the ratio of the total traffic cache and forwarded packets to the node’s actual cache capacity. In scenarios without packet loss at the node, a lower load rate indicates that tasks are being forwarded to other nodes, thereby increasing node utilization and enhancing overall network performance. However, during network topology changes or disruptions, the cache of the node increases, leading to a rise in the node load rate as some tasks may not be promptly forwarded in time. Therefore, this paper introduces the average node load rate to assess the performance of the IoV mesh network.

$$S\_node=\frac{\sum \frac{{\mu }_i+L_i}{C_i}}{N}$$

(8)

The above equation is the average load rate of the nodes in the network, ${\mu }_i+L_i$ denotes the sum of the traffic cache and the number of forwarded packets of the vehicle node i, and $L_i$ is the cache length at the current moment after updating according to Section 3.3. $C_i$ indicates the actual caching capacity of the vehicle node i, and N represents the total number of nodes in the network.

4.4. Link Load Rate

During task transmission in a network, traffic flows through the network links. The ratio of the actual traffic transmitted by a link to the theoretical capacity of the link is defined as the average load rate of the link. This metric provides insight into the utilization of network links; a higher value under no packet loss signifies increased network link utilization. Consequently, this paper introduces the link load rate factor to assess the performance of the IoV mesh network.

$$S\_link=\frac{\sum \frac{{\Psi }_{i,j}}{R_{i,j}}}{E}$$

(9)

In the equation above, ${\Psi }_{i,j}$ represents the actual traffic rate of the link from i to j, $R_{i,j}$ represents the theoretical rate of the link from i to j, and E represents the total number of network links. When the network is disturbed, resulting in packet loss for some tasks, the load on some links decreases, thereby resulting in performance degradation of the network.

4.5. Total Network Traffic

The total network task traffic is the aggregate of forwarded task packets and cached task packets within the network. This indicator represents the overall task traffic handled by the IoV mesh network. When the network is congested or disturbed, the total network task traffic will increase rapidly in a short period of time. Therefore, this paper introduces the total network traffic to evaluate the performance of the IoV mesh network.

$$sumflow=\sum {\mu }_i+L_i$$

(10)

where ${\mu }_i+L_i$ denotes the sum of traffic caching and forwarding packets of vehicle node i. $L_i$ in Equation (10) is the cache length at the current moment after updating as in Equation (8).

5. Simulation Results

In order to verify the validity of the model proposed in this paper, this subsection performs numerous performance simulations on the IoV mesh network to explore its performance variations under different scenarios. The simulation platform adopts PyCharm, and the CPU of the simulation computer is Intel Core i5-1240P. The experiment uses the case shown in Figure 2. The experiment is set up with four RSUs and 20 vehicles which move uniformly in a linear fashion according to their travel directions.

The simulation is divided into 300 time slots, of which the first 100 time slots are for the upper and lower lanes with the red light on and the left and right vehicles driving through. The next 50 time slots are for yellow lights for the left and right lanes. The last 150 time slots have the red light on for the left and right lanes and the up and down vehicles move. Since the vehicle speed is low in the urban scenario, the effect of the acceleration and deceleration process on the topology change is negligible, and in order to simplify the system modeling, we simulate the vehicle in uniform linear motion in the moving state. The length of each time slot is 100 ms. Each time slot has a fixed number of vehicles and RSUs to generate tasks. Referring to [31,33,34,35,36,37,38,39], the rest of the simulation parameters are shown in

Table 1. System parameters.

Parameter	Value	Parameter	Value
V2V link bandwidth	20 MHz	V2I link bandwidth	40 MHz
Vehicle transmission power	100 mW	RSU transmission power	20 dBm
Background noise power	−100 dBm	Absolute vehicle speed	[20, 40] km/h
Maximum communication range of the vehicle	200 m	RSU coverage range	500 m
Size of vehicle cache	100 Mb	Size of RSU cache	500 Mb

. The maximum packet forwarding rate for vehicle nodes is 10 Gbps [40].

5.1. Network Performance for Different QoS

This subsection focuses on simulating the relationship between task traffic and network performance metrics. The QoS in the simulation graph represents the traffic of the node initiating the task at each moment. Figure 8 explores the fluctuation of network packet loss rate and task arrival rate under different QoS.

As shown in Figure 8, initially, the network traffic and task arrival rate exhibit an increasing trend and then change dynamically. This is because under the task QoS and link rate constraints, tasks usually cannot be sent all at once. Consequently, upon the initiation of a task, network traffic experiences a continuous ascent. Subsequently, task transmission persists over time, with new tasks entering the network periodically, thereby the traffic distribution on nodes and links is in a dynamic process.

When the QoS is 20 Mbps, the average packet loss rate is approximately 3% and the task arrival rate is around 80%. As the QoS increases, the packet loss rate gradually rises, while the task arrival rate declines. This trend stems from the escalating network ingress traffic resulting in the expansion of the node’s cache queue. When the cache queue is full, packet loss occurs. Moreover, the higher the QoS is, the easier it is for the cache queue to fill up. In addition, under the same QoS setting, the declining task arrival rate can be attributed to the movement of vehicles causing link disruptions with the change in time, and most of the packets accumulate at the nodes because they cannot be transmitted according to the original routes when the port routing table is not updated. Simultaneously, the reduction in task arrival rate is further influenced by continuous task initiation leading to heightened network traffic over time.

Figure 9 simulates the variations in the other three metrics with QoS. At lower QoS levels, both the link load ratio and total network traffic exhibit an upward trend with QoS. This is due to the fact that when the QoS is small, the network loses fewer packets, and the total traffic increases with the increase in ingress QoS. Simultaneously, the number of packets forwarded by the nodes increases with the rise in the total number of packets, and thus the link load rate rises. However, at higher QoS settings, the discrepancy between the link load rate and the total network traffic diminishes significantly. This occurrence results from the heightened network packet loss rates in this situation, leading to a small disparity between the number of packets that the nodes can forward after packet loss and the total number of packets carried by the network.

For the node load rate, the trend is similar to the other two indicators under higher QoS conditions. The reason for this is the same as analyzed above when a higher degree of packet loss occurs for both vehicles and RSUs. Conversely, under lower QoS settings, the trend in the early stage is similar to the other two indicators. However, in the later stage, the cache queue of vehicle nodes tends to reach capacity due to their smaller cache size compared to RSUs. At this time, a majority of packet losses in the network occur at vehicle nodes, while the RSUs are minimally affected. Therefore, the increased total network traffic primarily originates from RSUs, while the increased packet loss from the vehicles results in a decreased overall node load rate in the network at QoS = 60 Mbps compared to QoS = 20 Mbps.

5.2. Network Performance for Different Caching Capacities

From the point of view of the node’s performance, Figure 10 simulates the variation in network performance with different cache capacities. The initial vehicle and RSU cache capacities are 100 Mb and 500 Mb, respectively. This subsection reduces both by half and observes the results in conjunction with QoS. From the perspective of packet loss and task arrival rate, reducing the cache capacity by half has a minor effect on both metrics at low QoS levels, whereas the impact becomes more pronounced at higher QoS values. This is because, in the case of less network traffic, the lower cache capacity can carry the task packets efficiently. And, under high network traffic conditions, the increase in the total number of packets fills the lower cache queue quickly. Consequently, there is a notable surge in packet loss rate alongside a significant decrease in task arrival rate.

Regarding the link load rate, reducing the cache capacity by half at the same QoS level reduces this metric. This outcome stems from the decreased storage of packets on the node due to the diminished cache capacity, leading to a corresponding decrease in the number of packets forwarded by the node. Furthermore, the impact of altering cache capacity on the link load rate is more obvious at QoS = 100 Mbps compared to QoS = 20 Mbps, aligning with the rationale provided for the packet loss rate analysis. Analyzed from the perspective of total network traffic and node load rate, under equivalent QoS conditions, the disparity in total network traffic before and after cache capacity adjustment demonstrates a growing trend over time. Moreover, this divergence becomes more conspicuous with higher QoS values. The reason is that halving the cache capacity leads to a large number of packet losses, the total number of packets carried by the network decreases, and the decreasing trend is more obvious with the increase in traffic.

In addition, the reason for the trend in node load rate with QoS = 20 Mbps is similar to the analysis in Section 5.1. When QoS = 100 Mbps, the total network traffic decreases following a halving of the cache capacity, but the node load rate remains nearly unchanged at this time. This is because when the cache capacity is larger, although there are more packets, they are relatively evenly distributed across the nodes. In contrast, reducing the cache capacity results in vehicle nodes carrying larger packets, so that the overall node load rate of the network does not undergo a significant decrease compared to the previous one.

5.3. Network Performance for Different Vehicle Densities

Figure 11 simulates the variation in network performance with different numbers of vehicles. As the quantity and density of vehicles rise, the number of nodes initiating tasks increases correspondingly, leading to an overall enhancement in the network’s total QoS. Consequently, this results in heightened packet loss rates and diminished task arrival rates. However, when QoS = 60 Mbps, the increase in the number of vehicles has less impact on these two metrics. This phenomenon arises from the network’s capacity to efficiently accommodate a greater number of task-initiating nodes under lower ingress QoS levels. In contrast, at higher ingress QoS values, the rapid consumption of limited cache resources by the augmented total packet count hampers network efficiency. As a consequence, there is a notable surge in packet loss rates.

Analysis of Figure 11 indicates that prior to approximately 150 time slots, networks with a high number of vehicles at the same QoS exhibit a lower link load rate than those with fewer vehicles; however, post 150 time slots, the pattern reverses. This shift occurs because as vehicle density rises, the number of links in the network topology also increases correspondingly. According to Equation (9), this results in a decrease in the link load rate. Conversely, as time increases, the impact of an increase in the total number of packets in the network outweighs the effect of an increase in the number of links, leading to a higher volume of packets being forwarded by the nodes and consequently an escalation in the link load rate compared to periods of low vehicle density.

For the total network traffic metric, the value of the metric for QoS = 60 Mbps and Vehicle = 32 is higher in the early phase than QoS = 180 Mbps and Vehicle = 20; however, the values become nearly identical in the later phase. The reason is that in the early stage of the network, the network with high vehicle density has enough node caches and links to manage packet storage and forwarding. As time progresses and the node caches reach capacity, the disparity diminishes due to packet loss. The reason for the trend in node load rate is similar to the analysis in Section 5.1 and Section 5.2. When QoS = 60Mbps, the metric displays a decreasing trend in the later stage as the number of vehicles increases. According to Figure 11d and Equation (8), at this time, the total network traffic gap is smaller and the number of nodes is larger, leading to a decrease in the node load rate.

6. Conclusions

In this paper, we proposed a network performance analysis model based on actual switches for IoV mesh networks. Initially, a mathematical model was formulated based on a typical IoV mesh network architecture to obtain the evolving network topology in real-time. Subsequently, by delving into the ingress and egress traffic forwarding mechanisms of individual nodes and expanding them to a multi-node multi-task environment, we established the task generation and the traffic forwarding model using actual switches. By integrating the dynamic network topology, we derived the current traffic distribution within the network. This information was then utilized to extract pertinent parameters for constructing the IoV mesh network performance evaluation indicator system, offering a comprehensive and scientific reflection of the network’s condition from both the task and network perspective. Simulation results demonstrated that network performance undergoes different but regular changes under diverse ingress QoS levels, caching capacities, and vehicle densities. The proposed model provides a rational and effective means of assessing network performance under different traffic conditions.

In future research, we can study more complex dynamic network scenarios, including networks with diverse node types and multiple channels, based on this network performance analysis model. Additionally, we can also explore problems such as task offloading, resource allocation, and load-balanced routing design of the network. The effectiveness and reasonableness of the formulated strategies are assessed by single or multiple metrics.