Abstract
In last few years, the Internet of Things (IoT) platform, which comprises a significant number of devices equipped with sensors and utilized for monitoring and actuation operations, has seen significant growth. Massive volumes of data are collected by IoT devices, which are subsequently sent to the cloud for analysis and forecasting. IoT data that is directly sent to the cloud can congest IT infrastructure that is used for other purposes. As a result, data offloading has become a popular issue in academic and industrial sectors, specifically for traffic-intensive applications that can benefit from offloading to local edge and cloud infrastructure. A lot of in-depth studies have been undertaken in areas where sensor device mobility is high in order to ensure efficient data offloading. Nevertheless, very few studies have taken into account the issue of data offloading with sensor devices that have very limited mobility, like industrial IoT sensors, which really needs more attention as it is one of the most data-trafficking areas. To counter this problem, a data offloading-based heuristic technique using edge computing is proposed in IIoT-based applications. This will reduce data transferring quantity, resulting in lower data migration capacity, bandwidth usage, and load, lowering operational costs. It is being put through extensive testing to see how it compares to existing algorithms. The experimental results show that the proposed algorithm is superior in terms of energy consumption, dependability, and resilience under various situations.
1. Introduction
The expansion of the Internet of Things (IoT) is considered a major success. In coming years, billions of connected systems will be used in homes, offices, businesses establishments, hospitals, and other areas. The computation power of IoT devices, on the other hand, does not ensure that tasks are done on time. As a result, it has become normal practice to outsource jobs to network edges and process them there [1]. As a result, concepts like edge computing (EC) [2], mobile edge computing (MEC) [3], mobile cloud computing (MCC) [4], fog computing (FC) [5] and others have been introduced in recent years. In order to take use of the processing and storage power of edge devices, academics and business have stressed the interaction of edge networks and IoT devices [6, 7]. As a result, data offloading has become an important strategy for Internet of Things (IoT) applications, especially those that run on mobile-based devices. One of the most important methods for boosting the resource usage of low-power devices is offloading (e.g., CPU, battery, and storage). The process of a device outsourcing the processing of a task to a more powerful unit is known as offloading [7]. Nonetheless, few studies have looked at several methods to optimize data offloading strategies using edge and fog computing to reduce transmission delay, enhance the communication delays, and increase privacy entropy. Many AI systems require as much data as possible to improve their performance, and data is currently regarded one of the most valuable resources. Offloading data to the edge or cloud center has become a significant alternative for permanently keeping data due to the lack of dependability and security.
Full data offloading is still a hurdle in IoT networks. Smart application-based IoT devices, such as vehicular networks and UAVs, offload sensor data to a nearby device that can compute and process the data. Offloading all data to the edge server will help to safeguard the data’s privacy while also removing the latency factor for important and compute-intensive applications. A lot of researchers have proposed the offloading technique using edge computing for compute-intensive applications and have improved the efficiency of the frameworks, but with the exponential rise of data for traffic-intensive applications like IIoT, the performance of the data offloading technique will be degraded as full data offloaded to the edge server will reduce its efficiency. So, using partial data offloading will offload some data at the edge and some at the cloud server to reduce traffic load on the server as shown in Figure 1. In the previous research conducted so far, data offloading techniques are used in IoT networks where the mobility of the nodes is high and the processing power of the device is low [8]. This has improved the efficiency of the mobile nodes, but with more and more devices being connected to the application, it will generate more traffic and this will degrade the data offloading performance. As a result, partial data offloading is regarded an efficient strategy for reducing scalability issues and energy usage while improving QoS. Nonetheless, few studies have looked at several methods to optimize data offloading strategies using edge and fog computing to reduce transmission delay, enhance communication delays, and increase privacy entropy. For the aforementioned difficulties, concentrate on high-reliability partial data offloading while reducing energy usage and traffic. So, develop a heuristic-based rate controller scheduling algorithm (HE2DRC), based on the partial data offloading model, to reduce traffic and energy consumption as much as feasible while maintaining data offloading dependability. The main contributions of this paper are as follows. (i)Offer an energy-efficient partial data offloading architecture for IIoT-based applications. This structure aids in traffic reduction and energy conservation. On edge-based IoT networks, it is utilized for partial offloading(ii)This is an innovative work towards curbing the traffic generated by sensors using partial data offloading under static devise, which presented a heuristic approach that can cut energy use while also avoiding traffic congestion(iii)Conduct an extensive experiment on real test beds. The suggested algorithm’s dependability, efficacy, and robustness are all proven when compared to previous algorithms
Further, the paper is presented in sections. Section 2 contains related work on data offloading. In Section 3, the model of the system and problem formulation are discussed. In Section 4, a heuristic selective offloading-based data rate controller algorithm is developed for industrial-based smart sensors in edge computing. Section 5 presents a real-time experiment using Arduino microcontroller and comparative analysis. Finally, in Section 6 are outlined the conclusion and future work.
2. Related Work
Researches on offloading concept in IoT have improved the resource usage of low-powered devices like CPU, battery, and storage. In [9], the authors proposed a unique decision-making paradigm for data offloading in which users can partially offload their data to multiaccess edge computing. Servers are installed on the ground and on UAVs. Zhang et al. suggested a DRL-based offloading technique in [10] using intelligent reflecting surface- (IRS-) aided MEC systems. Combination of IRS with MEC using DRL is able to improve performance of wireless transmission system. In [11], authors have shown the optimization procedure to derive the proper number of active processors allocated at each edge computing node, fog computing node, and cloud. In [12], an SBS offloading based Q-learning ranking is proposed for IoT devices. This approach is implemented in SDN controller to control offloading. The proposed approach aids in load balancing amongst SBS and reduces data communication lag. A 3-layer fog framework for urban IoT network is described in [13], in which a central service server employs a heuristic-based algorithm to dynamically request critical sensing data from specific IoT sensors from appropriate mobile gateways. In [14], by merging fundamental principles and methodologies from game theory and reinforcement learning, an AI-driven data offloading solution was proposed by Georgios et al. that enable UAVs to efficiently offload some of their data to a collection of MEC servers for additional processing. Overall, the framework is efficient and successful in a range of environments. The authors in [15, 16] have worked to improve the privacy of the offloaded data on edge servers using evolutionary algorithm SPEA2. In [17], Huang et al. proposed a data offloading model using SDN controller to minimize the traffic between vehicles and the base stations to ensure all the tasks are done within time constraints. Results showed that offloading strategy can greatly reduce the vehicular cellular traffic. In [18], Ghosh et al. proposed a 3-prediction-based offloading scheme that exploits the mobility pattern and temporal contacts of node to predict future data transfer options. After tests conducted, the proposed three systems show significant improvements in performance. Xu et al. suggested a heuristic offloading strategy for deep learning edge services in a 5G network in [19], with the goal of identifying the best destination VMs from edge computing nodes and cloud to reduce offloading time. In [20], the author proposed offloading data using fog in urban IoTs. The work focuses on optimizing communication in fog network. DEED, a novel dynamic energy efficient data offloading scheduling method, is developed in another paper to improve the energy efficiency [21, 22].
In literature survey, most of the work on data offloading has been done to improve the resource usage of low-powered devices with high mobility like UAVs, vehicular networks by offloading full data to the nearby devices having high processing power using cellular base stations. This has improved the efficiency of smart IoT devices, but researchers have not considered the consequences of offloading data to the high-powered device. This would increase the traffic on the devices with high processing power and will lead to congestion due to exponential growth of IoT-based smart devices. Thus, this paper uses the partial data offloading technique on devices with less mobility like industrial IoT smart machines which generate data at a very rapid pace to address the traffic problem while minimizing the bandwidth and load on the local and cloud server.
3. System Model and Problem Formulation
The overall system’s state transition as well as the problem formulation is discussed in this part.
3.1. System Model
Regarding data loads and network congestion limits, especially in wireless networks, full sensor data rates cannot always be supported when wireless temperature and humidity-based sensors like the DHT11 are employed in heating systems of the industry as shown in Figure 2. Edge computing solutions keep processing on the local server and can assist alleviate network congestion. It is presumed that the data will be processed by the local edge server or cloud when using the DHT11 sensor to gather temperature and humidity data because the sensor DHT11 has not had sufficient processing capability given the cost, performance, and battery constraints. As a result, it is assumed that the sensor could only identify the temperature and humidity of the boiler and that the local server can take immediate action if the temperature falls below the threshold. Sensor data (temperature and humidity) is only transferred to the cloud for enhanced processing if the local server recognizes a temperature beyond a threshold. It is not necessary to send data to the local edge server or the cloud when sensor data is within normal range. If the DHT11 sensor data falls outside of the typical range, it should be communicated to the local server for immediate action by the user if the value falls below a threshold. The DHT11 sensor must then transfer the data to the cloud if the local server identifies the data as being above the threshold value. Considering the properties of collected data, Figure 3 depicts system state changes. In the idle condition, the system considers the data to be within normal limits. The status is changed to an investigative state if sensor data is below or over the threshold limit. If the data collected falls below the threshold, the local edge server takes immediate action. If the amount of data exceeds the threshold, the system is switched to active mode, and the data is delivered to the cloud via the local edge server. If the sensor data returns to normal, the system state from both the investigation and active modes is restored to idle. The activity for each state is listed in Table 2.
3.2. Problem Formulation
In data offloading, resource-deficit devices offload their data to the resource-rich devices for computation, storage, and battery life. But with the rapid expansion of IoT devices, offloading every time to the nearby devices is not an efficient method as it will overburden the devices and increase load and bandwidth consumption will increase. The challenge addressed in this research is to use a partial data offloading approach in an IoT network based on edge computing to reduce data traffic in the industrial IoT sector. To put it another way, the ultimate goal is to control IoT traffic. As a result, the problem formulation is as follows: reduce (traffic) to a bare minimum.
4. Data Offloading Algorithm
4.1. Method Overview
Sensor data generated by smart devices is traditionally sent to a cloud platform. The rapid proliferation of IoT-based devices will result in an increase in data migration volume at the cloud end, leading to an increase in data uploading volume, consuming bandwidth, and increasing latency. In our method, DHT11 sensor data will be collected in Firebase which is used to collect real-time data. With the help of Google Cloud Platform (GCP), stored files in Firebase can be easily accessed by other projects in GCP, and using inbuilt metrics support, we plotted the graphs as per our need, imported the JSON files, and plotted the graph in Excel. This algorithm will use upper and lower threshold values to segregate the data. Upper and lower threshold values will be set as per the use case.
| ||||||||||||||||||||||||||||||
5. Experiments
A series of real-time observations on real testbeds are used to evaluate the performance of a heuristic-based energy-efficient data rate controller. In our tests, the proposed algorithm’s performance will be evaluated in two scenarios. Without using the proposed method, one scenario will dump all data to the cloud. The collected results will be compared to the outcomes of the second scenario, in which data is offloaded to a cloud server utilizing the proposed algorithm.
5.1. Experimental Setup
In this experiment, UDAC board is used, which contains an Atmega microcontroller and a DHT11 sensor containing temperature and humidity sensors with calibrated digital output and node MCU (ESP8266) are integrated together as shown in Figure 4. Wi-Fi connectivity is provided by node MCU, and sensor data can be sent to Google Firebase for analysis via microcontroller Atmega 256. The code is written in Arduino software and requires libraries to compile the code, which include libraries for Firebase ESP8266 to establish a Firebase real-time database connection with node MCU, DHT11 sensor library for reading the data from the DHT11 temperature and humidity sensor. The written code is compiled and uploaded on Atmega 256 microcontroller, and then, the sensor data is uploaded on the Firebase server for analysis using ESP8266. Table 3 presents the descripted value of the parameters used in the experimentation.
5.2. Performance Analysis of HE2DRC
The goal of HE2DRC is to reduce traffic by lowering data uploading volumes, decreasing data migration volumes, lowering bandwidth usage, and minimizing load, all while saving operational costs.
In the graph representation in Figure 5, the upper and lower threshold values are considered as 25- and 20-degree centigrades, respectively. As per the algorithm, if temperature data remains in between the upper and lower threshold value, data will not be uploaded, and if the data is below or above the thresholds, then temperature data will be offloaded to the cloud for further analytics.
5.3. Comparison Analysis
Here, Figure 6 shows full data offloading on server under normal conditions where all the data collected from the sensor is moved to the cloud server directly. The data collected is compared with another scenario, where partial data is offloaded on cloud server using heuristic-based algorithm. The result obtained shows a significant improvement in curtailing traffic by reducing the amount of data transferred at the cloud server end. Here, Figure 7 shows the amount of data stored in the server under two scenarios. Data stored under normal scenario, i.e., with full data offloading stores data and increases the data volume at the cloud server exponentially as compared with partial data offloading where rate of volume increased is managed with the help of the proposed algorithm. The results show that the proposed algorithm is effective in managing the increase of data volume. Figure 8 shows mean data load on server under two scenarios. Results shown in graphical format show the effectiveness of the proposed heuristic algorithm as load is reduced compared to the existing scenario. Figure 9 shows bandwidth consumed by server under two scenarios. In normal scenario, bytes sent per unit time are more compared to bytes sent under partial data offloading scenario.
Table 4 provides the overall improvements in the result after using the proposed partial data offloading. Amount of data offloaded and data stored at the cloud end shows significant improvements with 77.1% and 75%, respectively. Similarly, bandwidth consumption also shows improvements with 67%. Despite improvements in mean data load on server under partial data offloading, there is a scope of improvement in results which will be taken in the future work.
6. Conclusion
The paper focuses on selective data offloading with static mobility of the device using edge computing to reduce data traffic in the IoT network. For this, a heuristic-based energy-efficient data rate controller (HE2DRC) is proposed which is capable of handling the problem of data traffic in IoT networks. The paper is the first of its kind, where the data offloading concept is implemented on static smart devices rather than mobile-based smart devices. Real-time experiments were carried out, and data collected was graphically presented to ensure the effectiveness of the proposed model. For future work, offloaded data after segregation will be allotted tasks (local and cloud server) as per the task type using schedular algorithm, and finally, the efficiency of the overall model will be evaluated using different parameters like bandwidth, latency, and traffic rate.
Data Availability
No data were used to support this study.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this article.