SDN provides a unified model of the forwarding plane that is separated from the control plane. Meanwhile, NFV concentrates on visualization.

The use of SDN/NFV on edge nodes leads to reduced latency and increased scalability.

The rest of this paper is organized as follows. Section 2 presents a detailed literature review. Section 3 explains the proposed architecture and indicates its simulation. Section 4 discusses the simulation result and comparative analysis with existing architectures. Section 5 concludes this work.

With the advancement in IoT, large volumes of data are produced from IoT devices, such as smart home appliances, tablets, smartphones, routers, smart traffic systems, smart energy meters, cellular base stations, connected vehicles, and smart building controllers. Cloud computing efficiently processes considerable data in networks. With the rapid growth of the kind of IoT devices, excessive amounts of data must be transferred continuously over the network within a particular time. These data must be processed and responded as early as possible. As a result, the edge devices in the network are near to users, and the latency for transmitting data from the cloud data center to users is reduced, which leads to fast computation. We redevelop different communication methods with fog computing, such as SDN, NFV, content distribution network, the concept of 5G technology, and Long-Reach Passive Optical Network.

An overview of the proposed work is presented in Fig. 1. Cloud services are stored at the top of the layer called tier 3, which consists of the cloud server with large capacity. Tier 2 consists of an edge node, which is connected to all other networks and comprises all network devices, such as routers and gateways. The proposed SDN/NFV architecture is applied on the edge node. Tier 3 consists of IoT devices, such as cameras, smartphones, sensors, and computers.

Fig. 2 shows a detailed overview of our proposed architecture. It consists of three layers: application, control, and infrastructure. The first layer is responsible for network services at the cloud data center. These services communicate to the control plane of SDN's second layer called control layer through the open SDN orchestrator. In this proposed work, an SDN controller called Ryu is used. Ryu is an open-source structure created by NTT and written in Python. It supports OpenFlow, OF-Cong, and NETCONF. SDN's control plane is in charge of the traffic control in the application layer. The edge nodes directly communicate with the end devices. They communicate with the application layer through the SDN controller. Accordingly, the latency will be reduced, and data processing will be increased. SDN's main feature is the division of the control and forwarding planes.

In Fig. 3, in SDN's control plane, software called SDN controller is installed, which is responsible for handling the network traffic. It will run on COTS systems. The data plane is responsible for transferring data based on rules. The SDN controller combined with NFV VNFs at the edge is responsible for managing the rules through API. It does not have direct access to API. Consequently, the scalability and data processing accelerate. This work mainly focuses on traffic control by monitoring the data transferred to IoT devices from clouds and vice versa.

The transferred data must be related to the query created by the end user, which means data processing is improved in terms of time. The task of transferring the correct data to the user is based on three indexes: i) higher, ii) lower, and iii) neutral. If the importance of a real-time task is high and the computational necessity is low, then the task belongs to the higher index. By contrast, when the importance of a real-time task is low and the computational necessity is high, then the task belongs to the lower index. That is, the smaller the calculation amounts are, the lower the complexity of the task will be.

   1. Data transmission in the industrial layer consists of IoT devices. A set of N(d) devices called N(d) = {1, 2, …N} exists, and each device has a task to be completed in a certain limit.

   2. The orchestrator layer in the proposed SDNFV is responsible for handling real-time tasks by using an edge node server through the SDN controller via four actions: task creation, modification, operation, and completion.

   If request == valid then

        If task == exists then

             Executed from the catalogue of the network system

       Else

             Call the service-chained module of SDNFV

             Update the task

       end

   3. Once the task is created, activated, and completed, the computational complexity is calculated as(1) Tn={Sn,Cn,Ln,max}n∈N where

T_n − task for a particular device

C_n − calculation of the CPU cycle amount needed to complete the task

L_n,max − maximum latency of device n (in s)

   4. After all input D_n is received, the task is performed on the fog node. Thus, on the edge node, the execution time of task J_n consists of two sectors: transmission and computing time. f_n,E denotes the computational resource of the edge server.(2) Jn,E=Dnrn,E+Cnfn′E′

   5. The application layer performs the computation task based on the five indexes indicated in Eqs. (3)–(7) to analyze the cost, time, and computation.

The orchestrator layer in the proposed SDNFV is responsible for handling real-time tasks by using an edge node server through the SDN controller via four actions: task creation, modification, operation, and task completion. The workflow of the proposed architecture is depicted in Fig. 4.

From Fig. 4, the workflow of the proposed architecture is elucidated as follows:i)

Task Request: An IoT device (user) sends the task request to the cloud processor. It is described using predefined templates.

The proposed SDNFV model considers five indexes: cost, time, bandwidth, energy consumption, and energy calculation.i)

Cost Index

The cost can be calculated using Eq. (3). The cost of the distance from source to destination is calculated by multiplying W_c as the unit of communication cost with the distance. The lower the cost is, the better the architecture will be.(3) C(Si,Dj)=Wc×dsi,dj

ii) Time Index

It is the time the user needs from sending data to receiving data from the cloud center. Minimum time leads to a better model. It is calculated by adding transmission and computational delays as(4) T=Tp+Tc

iii) Bandwidth Index

m virtual machines exist at the cloud center to serve m users at a time, then the total bandwidth is fixed, as shown as follows [1]:(5) B={Bmax,m<T1W/m,T1≤m≤T2Bmm,n>T2 where B is the amount of bandwidth, and T₁and T₂ are the numbers of users at the cloud center.

iv) Energy Consumption

Energy consumption is the energy spent when the cloud center offers a service to device users. This is also called cost index, which means lower energy leads to a better model.

(6) Esi=Eb+m∗Cvm where E_ai is the total energy consumed, E_b is the total standby power consumption, and c_vm represents the unit of energy consumption.v)

v) Capacity of the Machines at the Cloud Center

The capacity is calculated as(7) Energy capacity = total energy capacity of the machine/ number of users

To evaluate the performance of the proposed architecture of SDNFV on edge nodes, the simulation results are analyzed and compared with the results of existing architectures. The effective performance of the proposed architecture is shown in Fig. 5. Our proposed SDNFV architecture on edge nodes is the middle architecture between the IoT devices and cloud data center. While using the proposed architecture on edge nodes for data processing, the response accelerates and the throughput is reduced because the configuration and processing of data are performed on the edge server.

MATLAB is used to develop the simulation environment for algorithm execution. A system of IIoT with SDN is established on edge devices. The EstiNet simulator is used for the simulation. In SDN, OpenFlow is processed between a controller and a switch. The simulated OpenFlow with SDN controller enables an Ethernet switch to support control panels as in-band or out-of-band by using multiple controllers. To control the simulated OpenFlow switches, open-source controller software can be used in the simulated SDN. The SDN Wi-Fi infrastructure can be simulated using OpenFlow-enabled access and switch devices. This simulation platform is best suitable for researchers in SDN.

The proposed architecture lies on edge nodes near IoT devices. The response from the cloud center for a task request is processed with high execution time and low time delay, and the request is validated before processing. Hence, the user will obtain a response with low computational power and high performance. While the device count increases the time, the complexity throughput is also increased gradually. The simulation result is shown in Tab. 1. Figs. 6–8 show the simulation result of the proposed SDNFV architecture in terms of the number of IoT device requests, such as 50, 80, 100, 150, and 200. Our proposed architecture presents computational complexity of 1500 J, cost of 240, execution time of 230 s, and throughput of 2200 kbps for 200 devices. To analyze the performance of the proposed architecture further, it is compared with existing architectures, such as ASTP and SuVMF.

Total Time Delay: The total time delay of the architectures is shown in Fig. 9. As the device amount increases, the delay time is also increased for all the architectures. However, our proposed SDNFV has less total time delay compared with the other existing architectures because it selects each device task through the edge node of the framework via NFV. For the requests of 200 IoT devices, the time delay of the proposed architecture is 1800 s only, that of ASTP is 2500 s, and that of SuVMF is 3000 s.

Reliability. This is an important index for evaluating the success of any architecture in task processing. When the failure rate to complete a task is low, it means the reliability of the architecture is high. From Fig. 10. with the increase in the device number, the reliability of all architectures decreases. With the concept of SDN with NFV on edge nodes, the device requests are validated, and response is performed through VNF's forwarding plane in the orchestrator layer. The reliability of the proposed architecture is higher than that of the other existing architectures. The proposed architecture achieves 90% reliability on 200 devices, which is higher than that of the two other architectures (85% for ASTP and 70% for SuVMF).

Satisfaction. It is a comprehensive index for real-time systems. The best architecture must rapidly respond to requests and achieve user satisfaction. When the number of device requests is increased, the tasks must be processed rapidly; if the system is not able to respond on time, the satisfaction will be low. Fig. 11. shows the analysis of the architectures in terms of user satisfaction.

The proposed SDNFV on edge nodes has the highest satisfaction among the architectures. The proposed architecture lies on the edge nodes near the IoT devices. The response from the cloud center for a task request is processed with high execution time and low time delay, and the request is validated before processing. Hence, the user will receive the response as early as possible with full satisfaction. Our proposed architecture achieves 90% satisfaction compared with ASTP at 82% and SuVMF at 68%. In sum, our proposed SDN with NFV on edge nodes performs better in terms of total time delay, reliability, and satisfaction.