Contents Introduction
With the advent of 5G, in order to satisfy the massive data demand, the energy consumption of wireless communication network has also increased sharply when paying attention to the improvement on data transmission rate [1]. By 2020, there will be 50 billion connected devices in the world [2]. In addition, FIGURE 1 shows cisco’s average growth forecast for mobile data traffic around the world. As we can see from FIGURE 1, mobile data traffic will increase sevenfold from 2016 to 2021, reaching 48270 PB per month [3]. However, the explosive growth of data demand forced the increase in data service rate, and making energy consumption one of the most concerned points in the wireless communication network in the near future. It is conceivable that high energy consumption will leave the entire network in a low energy state, which in turn increase delay and decrease transmission efficiency. So energy efficiency has become one of the most important performance markers in the next generation 5G heterogeneous network [4], [5].
Fortunately, MDs carried by users have different similarity factors, such as residence, work place and interest. We can use the similarity factor between MD to complete effective data transmission to improve energy efficiency. However, the physical size of MDs tends to be small in order to meet the compactness and portability of MDs, so the battery capacity and processor performance of MDs in wireless communication networks are limited [6]. However, with the rapid growth of mobile data traffic in wireless communication networks, the number of data transmission and the data volume per transmission in a network are also greatly improved, but MDs do not have enough capacity to transmit a large amount of data due to limited battery capacity and processor performance. Also, MDs need not only to transmit data, but also to calculate some tasks, so MDs use more energy. And at present, many compute-intensive mobile applications are deployed on MDs, so the time for MDs to enter the sleep state is faster because of low energy [7]. As a result, the issue of energy efficiency becomes more serious and challenging in wireless communication networks. Fortunately, Gartner predicts that more than 250 million smart cars will be connected to high-tech networks by 2020. At the same time, a Nvidia px2 self-driving car has the same computing power as 150 MacBook pro’s. If vehicles can be utilized, not only the idle vehicle resources can be used reasonably, but also the cost of physical infrastructure can be greatly reduced [8], [9]. In a word, to improve energy efficiency, we start from reducing energy consumption of MDs. Especially, in order to reduce the energy consumption of MDs in wireless communication networks, we mainly study from the following two aspects.
On the one hand, we propose the MEC-based OMSNs. OMSNs derive from the end-to-end communication of mobile devices carried by people based on the encounter opportunity [10], [11]. When the transmission domain between users cannot be reached, OMSN usually adopt the store-and-forward method to complete the data transmission process [12], [13], and the message carrying MD forwards the message to a certain relay MD or the destination MD by means of the encounter opportunity caused by movement to forward. In wireless networks, MDs in an unpredictable geographical location need to communicate with each other at unpredictable geographical intervals. Therefore, as a multi-hop wireless technology, OMSN proposes that end-to-end data transmission can be achieved through “opportunistic communication”. Such communication relies on the movement of MDs and the effective routing-forwarding algorithm, so as to give a MD the opportunity to act as a relay MD in some end-to-end data transmission. In addition, an effective routing-forwarding algorithm is to select reliable relay MDs for efficient data transmission [14]. Compare to other routing-forwarding algorithms, an effective routing-forwarding algorithm have the fewer number of hops in all successful message forwarding processes, and then the less energy and delay that MDs spend in the forwarding processes. Therefore, how to select reliable relay MDs is a pretty important problem in a routing-forwarding algorithm?
On the other hand, besides the effective routing-forwarding algorithm, we innovatively apply MEC to 5G OMSNs. MEC is a promising technology that allows users to access the computing services of small servers distributed close to users, which are defined as edge clouds in MEC [6], [18]. Moreover, MEC differs from cloud computing in that its computing resources are distributed around the network and are close to the end users. As a result, the energy consumption and delay during communication are reduced when MDs carried by users use the computing resources of edge cloud [19]. Generally, edge cloud provides computing and caching services for MDs in the network. Hence, MDs with limited performance offload some computationally intensive tasks to edge cloud and use the strong computing power of edge cloud relative to MDs to process tasks. Alternatively, MDs directly cache tasks that need to be processed in MDs frequently into edge cloud [20]. In wireless communication networks, the forwarding of messages between MDs is extremely frequent. Therefore, it is the best choice to cache the routing-forwarding algorithm directly on the edge cloud compared with offloading a task before each next hop optimal relay MD is selected.
To sum up, in order to solve the problem of delay, especially high energy consumption in 5G network, and improve transmission efficiency and energy efficiency of the whole network, this paper proposes the FRRF algorithm based on MDs similarity in a MEC-based OMSN. The FRRF algorithm uses the historical movement information and social attributes of MDs to determine the similarity between MDs. Based on fuzzy reasoning system and information entropy, the transmission priority value is calculated to comprehensively evaluate the movement and social similarity between MDs and destination MDs. By comparing the transmission priority value obtained, the message carrying MD can forward the message to the MD with higher transmission priority value, that is, the message carrying MD provides the opportunity for a MD with higher transmission priority value to act as a relay MD. The main contributions of this paper are as follows:
We innovatively apply MEC to the OMSNs in 5G to relieve the computing pressure of MDs in the process of routing and forwarding [18], [21]. Moreover, we proposed the FRRF algorithm in the MEC-based OMSN. For the FRRF algorithm, based on fuzzy reasoning system and information entropy, the car takes both the MD movement and social similarity into account to determines the transmission priority value between MDs, and compares the transmission priority value between MDs in the network to select the next hop optimal relay MD.
In order to accurately determine the transmission relationship between MDs in the MEC-based system, the following operations within the system will be performed. First, MDs collect and update their respective status information by means of the encounter between two MDs, and then offload their respective status information to the corresponding nearest car, so that the car can use the FRRF algorithm to accurately calculate the comprehensive similarity between MDs to determine the special transmission relationship between MDs.
Simulation experiments simulate the MEC-based OMSN. The simulation results show the effectiveness of the proposed FRRF algorithm. More specifically, the simulation results show that the FRRF algorithm improves the delivery ratio of the entire wireless network, and reduces overhead on average, average end-to-end delay and energy consumption of the entire wireless network.
Related Works
In recent years, the academic has done a lot of research around the routing-forwarding algorithm in OMSNs [11], [16], [22]–[24], and proposed different effective methods under different application scenarios. In OMSNs, routing-forwarding algorithms are usually divided into two types: context-aware routing-forwarding algorithms and non-context-aware routing-forwarding algorithms. Context-aware routing-forwarding algorithm based on the similarity of nodes to select relay nodes through the social relations between nodes and the contextual information related to nodes [16], [22]. In addition, although context-aware routing-forwarding algorithms can improve the transmission environment and improve transmission efficiency, these algorithms usually need to manage a large amount of information and perform heavy computing tasks, thus bringing additional delay and energy consumption to the network. However, non-context-aware routing-forwarding algorithms perform flooding transmission, which brings many redundant message group copies to the network, and eventually leads to extremely high forwarding delay and energy consumption of the network [11], [23], [24]. It can be seen that both the context-aware routing-forwarding algorithm and the non-context-aware routing-forwarding algorithm will bring some extra delay and energy consumption to the entire wireless network, especially the non-context-aware routing-forwarding algorithm. Therefore, edge cloud is applied to OMSN to reduce the delay and energy consumption of nodes in the network. In this paper, we mainly studies the context-aware routing-forwarding algorithm based on node similarity in the MEC-based OMSN. Next, we will discuss some of the newer context-aware routing-forwarding algorithms, some of the context-aware routing-forwarding algorithms that involve mathematical methods, and the latest research in MEC.
In context-aware routing-forwarding algorithms, many studies calculate the similar level between nodes to define the relationship between nodes, such as the possibility of a future encounter between nodes, the moving trajectory of nodes, and community partitioning of nodes. Wang et al. [25] innovatively extracted social identity from messages generated by mobile nodes, and proposed the SlaOR algorithm that takes into account the multiple social identities of mobile nodes and their corresponding social influences. By the final simulation results, the performance of data transmission can be improved by taking social identity into account. However, the SlaOR algorithm does not consider a variety of social attributes. Wu et al. [26] proposed the SRBRA algorithm, which is based on social relations. Firstly, real-time data generated by nodes are analyzed and summarized, and then specific factors affecting social relations between nodes are extracted to calculate the value of social relations between nodes. Finally, according to the social relation value between nodes, the social relation value between the neighbor node and the destination node is sorted to select the optimal next-hop relay node to complete the transmission of messages. However, the SRBRA algorithm does not take the mobility of nodes into consideration. Besides, Mayer et al. [27] studied a framework that takes individual context, society and relationships as matching opportunity predictors. The proposed algorithm based on a series of studies can predict the cooperation opportunities of data transmission between nodes, and then determine the end-to-end communication between nodes in the network according to the cooperation opportunities between nodes.
Some mathematical methods and models are usually used in context-aware routing-forwarding algorithms, such as markov decision model, set theory and graph theory. Two of the next three papers use game theory and one uses graph theory. Nguyen and Nahrstedt [28] proposed a new context routing protocol (GT-ACR) based on game theory to select the most appropriate relay node to forward packets. Through the non-zero cooperation times of two nodes, the GT-ACR protocol builds the game depends on the context information, the distance between the corresponding node and the target node, and the encounter index. In [29], in order to determine the cost to achieve efficient data transmission, Talipov et al. designed a model based on user context replication and the graph theory, which is an online backpack problem. The scheme learns and predicts the context information of each node in order to calculate the data delivery probability of each node, and the number of copied messages is adjusted based on the given delivery threshold. However, the scheme only considers the data information in the process of node transmission, which means the decision accuracy of message transmission needs to be improved in the process of transmission. Besides, in [30], in order to find the vertex cover suitable for the perceptive tasks in the group, Phuong Nguyen and Klara Nahrstedt designed a new context-aware approximation algorithm. At the same time, in order to assign the sensor task to a more “socialized” device for better sensor coverage, a human-centered guidance strategy for initial assignment of the sensor device based on participants’ meta information was also designed. And the node of this algorithm completes the individual coverage of a social vertex with the human-centered information.
MEC has many research points, such as resource management on MEC, security and privacy issues in MEC, caching issues on MEC server, mobility management, deployment of MEC server, and green energy MEC. Some of the existing work uses MEC computing resources to relieve their own computing pressure [31]–[35]. In [31], in order to study the video collaborative processing scheme, Long C et al. proposed an edge computing framework for collaborative processing of delay-sensitive video tasks on MDs. The framework takes both video group matching and group information into consideration to maximize the detection accuracy of people in the video task deadline. In addition, both the execution delay of tasks on the camera and the delay of data offloaded to the edge server is reduced. In [32], under the application scenarios of augmented reality and other compute-intensive time-delay sensitive tasks, Chen M et al. proposed the architecture of the software defined super-dense network (SD-UDN). The devices under this architecture can offload tasks to the edge cloud. The main purpose of the paper is to solve the task offloading problem to minimize delay. Especially, the task offloading problem is a NP hard and mixed integer nonlinear programming problem, which is composed of task assignment sub-problem and resource allocation sub-problem. Li et al. [33] proposed new vehicle network structure in the smart city scenario, and combined optimization of cache, network and computing resources to alleviate congestion in the network. In addition, this structure introduces the programmable control principle of software-defined network. And after modeling the service, vehicle mobility and system state, the paper proposes the joint resource management scheme to minimize the system cost, namely the task execution time and network overhead, in which this scheme is a partially observable Markov decision-making process.
The rest of the paper is as follows: Section III shows the modeling process of the FRRF algorithm. Section IV shows the FRRF algorithm in detail and analyzes its performance. Section V provides simulation experiments to verify the theoretical analysis of the FRRF algorithm and its effectiveness. Finally, Section VI gives the conclusion of this paper. Some key mathematical notations are explained in Table 1.
System Model Design
As shown in FIGURE 2, we consider a system of MEC-based OMSN, which consists of
In the first stage, MDs in the network share network status information by encountering each other, so as to achieve the purpose of collecting as much MD information about the network as possible. The MD information here mainly refers to the movement and social preferences information of the MD.
In the second stage, message carrying MD offloads some information about the message (destination MD number) and MD information in the network collected by the MD itself to its nearest car through the uplink.
In the third stage, after receiving the information offloaded by the message carrying MD in the two stage, the car will take both the destination MD number of the message and MD information in the network as input to the FRRF algorithm. Then the car closest to the message carrying MD executes the FRRF algorithm based on the input data information. Finally, the optimal next hop MD number is obtained.
The fourth stage is that the car closest to the message carrying MD returns the execution result (the optimal next-hop MD number) of the FRRF algorithm to the message carrying MD by downlink.
In the fifth stage, the message carrying MD transfers the message according to the optimal next hop MD number received from the car.
A simple and successful message forwarding process in the mobile edge computing-based opportunistic mobile social networks.
In this section, we consider some steps of the process of message forwarding. The second, fourth and fifth stages mentioned above is beyond the scope of our research, and we mainly study the first stage and the FRRF algorithm involved in the third stage. Next, we mainly discuss how do MDs collect the network information and how do cars select the optimal next hop delay MD by the FRRF algorithm, that is car chooses a MD according to the FRRF algorithm and gives the MD the opportunity to act as a relay MD in the process of message forwarding. To avoid confusion and ease of understanding, the whole process of the message being transmitted from the start MD to the destination MD is defined as forwarding, in which the message transmission between MDs is briefly described as transmission.
A. Collect and Affload Network Status Information
In order to get the network status information of MDs, there exists a special preparatory period
Firstly, we quantify the process of collecting and updating MD movement and social habits during the preparatory period. We define a triple to denote the state of MD 
\begin{equation*} State_{n1} = (Dis_{n1},\alpha _{n1},List_{n1}),\tag{1}\end{equation*}
\begin{equation*} List_{n1}=\lt State_{a}, State_{b}, \ldots , State_{m} \gt,\tag{2}\end{equation*}
Meanwhile, during the preparation time \begin{align*} EState_{n1}^{1}=&State_{n1} \bigcup State_{n2} \tag{3}\\=&(Dis_{n1},\alpha _{n1},Dis_{n2},\alpha _{n2},List_{n1,n2}), \\ ES_{n1}=&(EState_{n1}^{1},EState_{n1}^{2},\ldots,EState_{n1}^{k}).\tag{4}\end{align*}
Secondly, we describe the process by which MD \begin{equation*} r_{n1}=B_{n1} log_{2}\left({1+\frac {P_{n1} h_{n1}}{w_{0}}}\right),\tag{5}\end{equation*}
After the message carrying MD (MD
Next, we will model the FRRF algorithm in the following two subsections.
B. Phase 1 of the FRRF Algorithm: Assess MD Similarity
Based on the collected information, the car in the MEC-based OMSN system will assess the movement and social similarity between MDs. We first assess the movement similarity between MDs.
In general, the more similar the moving trajectories between MDs, the greater the likelihood that the message will be forwarded successfully between MDs. Therefore, we use the moving trajectories between MDs to describe the movement similarity between MDs. Moreover, the moving trajectory of a MD based on spatial and temporal information of the MD, so the joint time point and communication area in the FRRF algorithm are used to describe the moving trajectories of MDs in the MEC-based OMSN system, such as \begin{equation*} Dis_{n1}= \sum _{j=1}^{J_{n1}} \frac {h(LS,p_{n1}^{j})}{J_{n1}}.\tag{6}\end{equation*}
In equation (6), if
In addition, we set different weights \begin{align*}&\hspace {-2.5pc}MS_{n1,n2} \\&=\,\frac {\sum _{j=1}^{J_{n1}} \sum _{i=1}^{J_{n2}} \theta (\Delta t - |t_{n1}^{j}-t_{n2}^{i}|) h(p_{n1}^{j},p_{n2}^{i})} {\sum _{j=1}^{J_{n1}} \sum _{i=1}^{J_{n2}} \theta (\Delta t - |t_{n1}^{j}-t_{n2}^{i}|)}.\tag{7}\end{align*}
Secondly, the social similarity between MDs is evaluated by studying the relationship between the social attributes of MDs in the network. The social attribute eigenvector of MD \begin{equation*} \alpha _{n1}=(A_{n1},B_{n1},\ldots,Y_{n1}),\tag{8}\end{equation*}
\begin{align*}&\hspace {-2.5pc}sim_{n1,n2}^{A} \\=&\frac {A_{n1} A_{n2} + \varphi }{max(\|A_{n1}\|^{2},\|A_{n2}\|^{2})+\varphi } \\&+\, \frac {min(\|A_{n1}\|^{2},\|A_{n2}\|^{2})-A_{n1} A_{n2}}{\|A_{max}\|^{2}},~(n1 \neq n2),\tag{9}\end{align*}
In addition, by evaluating the similarity of all social attributes sub-vectors between MD \begin{align*}&\hspace {-2.5pc}SS_{n1,n2} \\=&w_{1} sim_{n1,n2}^{A}+ w_{2} sim_{n1,n2}^{B} +,\ldots,+ w_{y} sim_{n1,n2}^{Y}, \\&(n1 \neq n2),\tag{10}\end{align*}
\begin{equation*} M= \left ({\begin{array}{ccccc} z_{11} &\quad \cdots &\quad z_{1v} &\quad \cdots &\quad z_{1y} \\ \vdots &\quad \ddots &\quad \vdots &\quad \ddots &\quad \vdots \\ z_{u1} &\quad \cdots &\quad z_{uv} &\quad \cdots &\quad z_{uy} \\ \vdots &\quad \ddots &\quad \vdots &\quad \ddots &\quad \vdots \\ z_{x1} &\quad \cdots &\quad z_{xv} &\quad \cdots &\quad z_{xy} \end{array} }\right)\tag{11}\end{equation*}
\begin{equation*} C_{uv} = \frac {z_{uv}}{\sum _{u=1}^{x} z_{uv}}.\tag{12}\end{equation*}
\begin{equation*} TC_{v} = - \frac {1}{lnx} \sum _{u=1}^{x} C_{uv} ln(C_{uv}).\tag{13}\end{equation*}
Thus, the weight of a sub-vector can be determined according to the corresponding contribution level. We now define the contribution consistency level of each MD on \begin{align*} w_{v}=&\frac {g_{v}}{\sum _{v=1}^{y}g_{v}}. \tag{14}\\ w_{v}^{*}=&\frac {\beta _{v} g_{v}}{\sum _{v=1}^{y} \beta _{v} g_{v}}.\tag{15}\end{align*}
In conclusion, based on the above research on the social similarity of MDs, the MDs in the MEC-based OMSN system can be divided into different communication communities. By taking advantage of the fact that MDs belonging to the same communication community have more opportunities to transmit data to each other, the success rate of message forwarding between MDs can be greatly improved.
Next, according to the fuzzy reasoning system, we will evaluate the transmission priority of MDs by combining the movement similarity and social similarity of MDs.
C. Phase 2 of the FRRF Algorithm: Compute Transmission Priority by the Fuzzy Reasoning System
In the FRRF algorithm, we use the fuzzy reasoning system to determine the movement similarity degree and social similarity degree between MDs, and then calculate the transmission priority between MDs. In fact, the special relationship between MDs can be determined by the similarity between MDs. However, some unstable factors between MDs affect the similarity between them. If the data transmission between MDs is determined based on the affected similarity, it will cause the MDs in the system to collect inaccurate network state information. Therefore, we do not directly use similarity between MDs to determine the data transmission between MDs. We first use similarity between MDs to obtain a transmission metric, which representing the fuzzy degree of message forwarding between MDs and can effectively avoid the disadvantages of the affected similarity between MDs. Moreover, because of the extensive applicability of Mamdani fuzzy system, we use Mamdani fuzzy system as the fuzzy reasoning system in our paper. The fuzzy reasoning system is composed of the fuzzifier, the fuzzy inference, and the defuzzifier, respectively. Next, we will describe the three components of the fuzzy reasoning system in detail.
Firstly, fuzzifier is used to compute the membership level of fuzzy sets in fuzzy reasoning systems. For the fuzzifier in our paper, movement similarity and social similarity are taken as two input variables, and three fuzzy sets are defined for both input variables, i.e., low, medium and high. Membership level of each fuzzy set can be calculated by corresponding membership function. Therefore, we define three different membership functions for the three fuzzy sets. In general, different membership functions are defined according to different scenarios, like triangular and trapezoidal. Since the movement of MDs in the system follows a normal distribution, we define a normal distribution membership function to evaluate the membership level of these two input variables in our fuzzy reasoning system, as shown below \begin{equation*} F_\mu (b) = \frac {1}{\sqrt {2\pi }\sigma } exp \left({{-\frac {(b-\mu)^{2}}{2 \sigma ^{2}}}}\right),\tag{16}\end{equation*}
Secondly, we do fuzzy reasoning. In the Mamdani fuzzy system, two input variables (movement similarity and social similarity) correspond to three different fuzzy sets, so the combination of the two input variables corresponds to nine different fuzzy sets, which is shown in Table 2. Based on the FRRF algorithm, we comprehensively consider the
It is easy to analyze from Table 2 that the movement similarity between MDs has a deeper impact on message forwarding than the social similarity between MDs. On the one hand, the reason is that the movement similarity reflects the similarity of the moving trajectories between MDs. Furthermore, the higher the movement similarity between MDs makes the two MDs more likely to meet, the greater the possibility of message transmission between MDs. On the other hand, social similarity divides MDs into different communities according to their social attributes. In detail, MDs with similar social attributes will be divided into the same communities. And the higher the possibility of transmitting messages between MDs belonging to the same community.
Finally, we show the third part of the fuzzy reasoning system after fuzzy reasoning, which is defuzzifier. Based on the Mamdani fuzzy system, the FRRF algorithm uses OR operation and AND operation to determine the transmission priority between MDs. In detail, the OR operation is first used to maximize the values of all fuzzy sets, followed by the AND operation to compute the minimum combination of the values of these fuzzy sets. In other words, the maximum shadow region of the value of each fuzzy set is obtained through the OR operation, and the minimum overlapping shadow region of the six largest shadows of the two input variables is obtained through the AND operation. Moreover, the maximum shadow region of the value of each fuzzy set represents the control result of each membership function. And data transmission priority can be obtained through the minimum overlapping shadow region of the six largest shadows of the two input variables, which is the data transmission recommendation results obtained by fuzzy reasoning based on movement similarity and social similarity. As for the final transmission priority value, the movement of the MDs in the system follows the normal distribution, so we use equation (17) to calculate the centroid of the overlapping shadow region and creatively use it as the transmission priority value.\begin{equation*} TPV_{n1,n2} = \frac {\sum _{l=1}^{L}F_{l} \cdot s_{l}}{\sum _{l=1}^{L}F_{l}},\tag{17}\end{equation*}
For the sake of understanding, we take an example and assume that the start MD (MD
The detailed process by which a message is forwarded from the start MD (MD
D. The FRRF Algorithm
In a word, the FRRF algorithm is a routing-forwarding algorithm based on fuzzy reasoning system to study the similarity between MDs in the MEC-based OMSNs system. The biggest difference from other routing-forwarding algorithms is that the FRRF algorithm is cached on cars with strong computing power in advance and executed on the cars. So using the computing power of the cars can save a lot of power for the MDs in the system. In order to better understand the FRRF algorithm, we have listed the detailed steps of the FRRF algorithm.
In the preparatory stage, each MD in the system collects movement and social attribute information, and builds its own state triple. In addition, each MD will share its own state triple and at the same time establish and improve its own encounter information matrix by the encounters between MDs.
The message carrying MD offloads the established encounter information matrix and the destination MD number of the carried message to the nearest car. According to the offloaded information, the car find the optimal next-hop delay MD, which has the highest movement or social similarity with the destination MD of the carried message. Because there is the highest movement or social similarity between the optimal next-hop relay MD and the destination MD, the optimal next-hop relay MD is more likely to be the relatives or friends of the destination MD, or have a highly similar movement trajectory to the destination MD. By selecting the optimal next-hop relay MDs in this way, the message can be successfully and efficiently forwarded from the start MD to the destination MD.
Based on the Mamdani reasoning system, the car computes the membership level of similarity between MDs. Also, the car determines the transmission priority level of these MDs versus the destination MD, and finally computes the transmission priority values between these MDs and destination MD.
The car determines the optimal transmission decision by comparing the calculated transmission priority values. And then the message carrying MD can transfer the message to the MD with the highest transmission priority value. By repeating the same four steps, the message can eventually be transmitted from the start MD to the destination MD.
Algorithm 1 The Proposed FRRF Algorithm
MD
/*The computation of transmission priority*/
MDs in the network collect information about all encounter MDs;
if
Give different weight for each sub-vector of the social attribute eigenvector;
Compute
for each membership function of the fuzzifier component do
Compute the membership degrees of movement and social similarities;
end for
Determine the transmission priority degree of MDs;
Output
end if
/* Forwarding messages */
if
if (
MD
end if
end if
More specifically, in the preparatory stage, the meeting MDs in the network share state sequences with each other to constantly update their encounter information matrix, hence the time complexity of this stage is
Simulation
A. Parameters Setting in Simulation
The simulation uses Matlab R2016a to simulate real scenario and evaluate the performance of the FRRF algorithm. In order to clearly show the advantages of the FRRF algorithm, we compare the FRRF algorithm with Epidemic [36], Spray and Wait [37], EIMST (Effective Information Transmission Based on Socialization Nodes) [38] and ICMT (Information Cache Management and Data Transmission Algorithm) [11], in which EIMST and ICMT are two new routing-forwarding algorithms, while Epidemic and Spray and Wait are two typical traditional routing-forwarding algorithms. In the simulation experiment, we set the relevant parameters as follows: The communication domain is a quare with
In addition to comparing the FRRF algorithm with the other four algorithms, the simulation focuses on four aspects, namely delivery ratio, overhead on average, average end-to-end delay and average remaining energy.
Delivery ratio: The parameter denotes the probability of choosing an optimal next-hop relay MD during the message transport phase. We define \begin{equation*} R = \frac {N^{rec}}{N^{sen}},\tag{18}\end{equation*}
Overhead on average: The parameter denotes the network overhead of successfully transmitting a message between MDs. We define \begin{equation*} O = \frac {T^{tot}-T^{suc}}{T^{tot}},\tag{19}\end{equation*}
Average end-to-end delay: The parameter consists of three parts, that is the delay of select the optimal next-hop delay MD, the delay of relay MD waiting for the message, and the delay of message forwarding. We define \begin{equation*} D^{ave}=\frac {D^{sum}}{\chi ^{suc}},\tag{20}\end{equation*}
Average remaining energy: The parameter denotes the average remaining energy of all MDs in the network at the end of the simulation experiment. Moreover, the energy consumption of the MD is composed of parts, that is energy consumption of basic operation when there is no computing task (including collect network status information), energy consumption of offloading encounter information and energy consumption of transmitting message. We define \begin{equation*} E^{ave}=\frac {E^{sum}}{\chi },\tag{21}\end{equation*}
B. Simulation Case Analysis
According to equation (16), for the low, medium and high fuzzy sets defined in our paper, we have three normal membership functions, i.e., \begin{align*} F_{1}(b)=&\frac {1}{\sqrt {2\pi }\sigma _{1}} exp \left({{-\frac {(b-\mu _{1})^{2}}{2 \sigma _{1} ^{2}}}}\right), \tag{22}\\ F_{2}(b)=&\frac {1}{\sqrt {2\pi }\sigma _{2}} exp \left({{-\frac {(b-\mu _{2})^{2}}{2 \sigma _{2} ^{2}}}}\right), \tag{23}\\ F_{3}(b)=&\frac {1}{\sqrt {2\pi }\sigma _{3}} exp \left({{-\frac {(b-\mu _{3})^{2}}{2 \sigma _{3} ^{2}}}}\right).\tag{24}\end{align*}
In normal distribution, three membership functions corresponding to three fuzzy sets, i.e., low, medium and high fuzzy set.
For example, when
The minimum overlapping shadow region of low, medium and high membership functions when
C. Result Analysis
Firstly, we show the performance of the FRRF algorithm versus the preparatory period
FIGURE 9 shows the delivery ratio
FIGURE 10 shows the overhead on average
FIGURE 11 shows the average end-to-end delay
FIGURE 12 shows average remaining energy
Secondly, we compare and analyze the FRRF algorithm with the other four algorithms. Since some algorithms in the OMSN are based on contextual information, MDs in the system need to carry, transmit and forward some text information about the network state. However, the limited cache space of MDs limits data transfer. Based on this, we set the buffer space size of the MD as a variable in the simulation experiment to study the transmission capacity of these algorithms. The experimental results show that compared with the other four algorithms, the FRRF algorithm performs better in delivery ratio, average end-to-end delay, network overhead and average remaining energy. Below, we compare the performance of the five algorithms in terms of delivery ratio, average end-to-end delay, network overhead and average remaining energy in detail according to FIGURE 13, 14, 15 and 16, respectively.
The average end-to-end delay
The average remaining energy
FIGURE 13 shows the delivery ratio
FIGURE 14 shows the overhead on average
FIGURE 15 shows the average end-to-end delay
FIGURE 16 shows the average remaining energy
Conclusion
In this paper, we apply MEC to OMSNs innovatively. In order to effectively reduce the energy consumption and delay of MDs in wireless network, we proposed the FRRF algorithm in the MEC-based OMSNs. The FRRF algorithm comprehensively considers the movement and social similarity between MDs to determine the transmission priority value between MDs, and finally make the optimal transmission decision by comparing the calculated transmission priority value between MDs. More specifically, the calculation of similarity between MDs is based on fuzzy reasoning system and information entropy. Further, the complexity analysis in the FRRF algorithm part shows the low complexity and high efficiency of the FRRF algorithm. Finally, the correctness of the theoretical analysis, the efficiency in reducing energy consumption and delay, and advantages over other algorithms are verified by the simulation results.