Thematic editorial: edge computing, fog computing, and internet of things

Edge computing processes data near the source, minimizing latency and reducing bandwidth consumption by avoiding unnecessary data transmission to cloud servers. This decentralized approach is beneficial for real-time applications such as augmented reality, autonomous vehicles, and industrial automation. Fog computing, on the other hand, extends the cloud closer to the network edge, creating a distributed layer of intermediate fog nodes that provide additional processing and storage capabilities. This architecture balances the low-latency benefits of edge computing and the resource scalability of cloud computing. In parallel, the rapid growth of the Internet of Things (IoT) has led to an unprecedented increase in data generation and connectivity requirements. As a result, traditional cloud computing models have faced significant challenges in handling the high data transmission, processing, and storage volume while maintaining low latency and high efficiency. Novel computing paradigms like edge computing and fog computing have become critical components in IoT ecosystems. This paper reviews recent developments on edge computing, fog computing, and IoT recently published in The Computer Journal. By analyzing this broad range of research, we will examine how these technologies complement each other and address the growing demands of modern networked environments. Edge computing refers to processing data closer to where they are generated, at the edge of the network. It reduces latency by performing computations locally rather than sending all data to the cloud. Fog computing, on the other hand, extends cloud computing closer to the edge, providing an intermediate layer between edge devices and the cloud. It distributes processing across multiple fog nodes to enhance performance. Finally, the IoT is a network of interconnected devices that collect, share, and act on data. These devices include sensors, cameras, and smart home appliances. IoT relies on edge and fog computing to manage the vast amount of data generated, ensuring efficient processing and timely decision-making. Edge computing offers the lowest latency, making it ideal for real-time applications. Fog computing provides more scalability by distributing processing power across intermediate nodes, though with slightly higher latency. IoT, on the other hand, depends on both edge and fog computing to process and manage the data generated by connected devices. While edge computing focuses on immediate data processing, fog computing balances the need for quick processing with cloud scalability, supporting larger, more complex networks. In this section, articles with a focus on edge computing are reviewed. Tang et al. [1] propose a data-cleaning method to improve object localization accuracy in edge computing environments. The authors address how to deal with noisy and incomplete data, a problem often encountered in edge-based localization systems. They introduce a hybrid data cleaning approach, combining statistical techniques with machine learning models to enhance data quality. The proposed method outperforms existing techniques by reducing localization errors and improving the reliability of edge-based systems. Simulation results demonstrate that the scheme can significantly enhance accuracy and efficiency in real-time applications, especially in dynamic and complex environments. Tian et al. [2] propose CCESHP, which is a causal consistency model using a hash ring structure and partial geo-replication, specifically designed for edge storage systems. The model addresses the challenge of maintaining consistency across distributed edge nodes, particularly with IoT devices. The key contribution is the use of a two-hash system to map data keys and servers on a hash ring, enabling efficient grouping and partial replication of data across edge nodes. This allows for lower latency and higher fault tolerance compared with full replication models. Additionally, CCESHP minimizes data storage and transmission overhead by only replicating essential subsets of data in geo-distributed regions. The model provides causal consistency, ensuring that operations on data respect the order of causal dependencies. Simulations demonstrate good performance in terms of consistency, latency, and system throughput, making it suitable for large-scale edge computing environments. Yang et al. [3] develop an analytical model to evaluate the performance of a heterogeneous edge data center (EDC), which is crucial for IoT applications. The model captures the dynamics of EDC systems that are subject to security threats and heterogeneous workloads. The authors focus on how attacks and failures impact the performance of these systems, which are essential for delay-sensitive IoT jobs. Using their model, they derive performance metrics such as system throughput and resource utilization. They also examine profit-related metrics for EDC administrators, providing insights on how to maximize profitability while maintaining performance under vulnerable conditions. Simulation experiments validate the accuracy of the model and suggest strategies for optimizing EDC performance and security. The results highlight the importance of balancing resource allocation and system protection in edge computing environments. Brahmi and Selmi [4] propose a trust-aware web services composition approach using a coordinate system in an edge and cloud environment. The method, called CWS_SMA, focuses on optimizing trust and quality of service by leveraging mathematical coordinates. This framework enhances web service composition by improving the selection of trustworthy services based on their spatial coordinates in a multidimensional space. Additionally, it introduces cooperative intelligent agents to reduce composition response time. The proposed system is evaluated for performance, showcasing significant improvements in response time and trust computation. The approach targets applications in environments like edge computing, where dynamic service compositions are essential for maintaining efficiency and security. Hsu [5] presents a computational offloading method based on a dueling Deep Q-Network (DQN) model for mobile edge computing (MEC) in the Industrial Internet of Things (IIoT). The proposed method addresses task offloading problems in MEC using reinforcement learning to optimize decision-making. The dueling DQN model is employed to enhance the efficiency of offloading tasks by separating the estimation of state value and action value, thus improving the learning process. The method reduces the latency of task processing while optimizing energy consumption in IIoT networks, critical for real-time industrial applications. Simulation results show that the proposed approach outperforms traditional DQN models in terms of task execution time, resource utilization, and energy efficiency. The article contributes to developing intelligent offloading methods in dynamic IIoT environments with limited resources. Yao et al. [6] address the optimization problem of model caching and request routing in edge-enabled wireless sensor networks (WSNs). The focus is on improving the efficiency of deep neural network (DNN) models under constraints like budget, latency, and accuracy. The proposed solution considers both loading cost and resource sharing in a collaborative edge environment. The problem is proven to be NP-hard, and an approximation algorithm based on randomized rounding is presented. The algorithm achieves a provable approximation ratio and demonstrates a 58.8% improvement in system throughput compared with baseline approaches. Extensive simulations validate the effectiveness of the proposed solution in improving model caching and routing performance in WSNs. Jiang et al. [7] present a model-based analysis of resource allocation policies in cloud-edge computing environments. The article explores various strategies for allocating resources between cloud and edge layers, with a focus on balancing computational load, reducing latency, and maximizing resource utilization. Through modeling and simulations, they demonstrate how different policies impact system performance under varying workloads. The study evaluates policies based on key metrics such as response time, energy consumption, and overall system throughput. It highlights the trade-offs involved in prioritizing either edge or cloud resources, offering insights into the best allocation strategies depending on specific application requirements. The article concludes that hybrid approaches, combining both edge and cloud resources, generally provide better performance for heterogeneous workloads. Zhang et al. [8] propose a framework for task offloading in MEC, aiming to balance energy consumption and task latency. The article proposes a collaborative task offloading scheme, where tasks are offloaded to edge servers, and computation results are reused to further improve efficiency. By leveraging partial offloading and result caching, the system reduces redundant computations, thus optimizing resource use. A game-theoretic approach is adopted to model user cooperation and task offloading, minimizing energy consumption. The article also explores the benefits of computation result sharing between users, reducing overall computational load. Extensive simulation results show that the proposed method significantly lowers latency and energy consumption compared with traditional offloading strategies. The framework is particularly effective in environments with dynamic and heterogeneous user demands. Pourian et al. [9] present a deep learning model for energy-aware task scheduling in fog computing environments. The focus is on reducing energy consumption, a key challenge in fog-based IoT systems. The proposed model uses a scheduling algorithm based on Learning Automata (LA) to optimize task scheduling for energy efficiency, makespan, and cost. The article further enhances the LA algorithm by integrating a neural network-based prediction model to forecast the relationship between makespan, energy consumption, and cost. The model successfully reduces the energy usage by 20% while maintaining efficient task scheduling. Extensive simulations validate the model’s accuracy and reliability, showing significant improvements over existing approaches. Hala and Sridevi [10] propose an approach to improve task processing in mobile real-time IoT applications within fog-cloud computing environments. The article addresses the challenges of mobility, security, and resource constraints in task scheduling. They introduce three algorithms to allocate mobile IoT devices to the appropriate edge device, considering factors like distance and bandwidth load. Additionally, a fuzzy-logic-based scheduling algorithm is developed to optimize task distribution between fog and cloud layers while ensuring security requirements are met. The proposed approach improves task processing time, turnaround time, and success ratio compared with existing methods. The results demonstrate that integrating distance and bandwidth considerations can enhance overall task processing efficiency. Xu et al. [11] propose a hybrid optimization algorithm for resource recommendation in fog-based IoT environments. The article focuses on improving resource allocation efficiency, by addressing the challenge of having unpredictable and highly dynamic fog environments. The authors develop a hybrid approach that combines cooperative filtering with the Artificial Bee Colony algorithm to enhance the accuracy of resource recommendations. Using CloudSim for simulation, the proposed method demonstrates improvements in accuracy by 1%–8% compared with traditional methods. Ma et al. [12] introduce KEFSAR, a solar-aware routing strategy for rechargeable IoT networks. The article addresses energy management challenges in WSNs powered by solar energy. KEFSAR uses high-accuracy prediction algorithms to optimize energy usage by taking into account the unpredictable nature of solar energy availability, such as weather changes and shadows. The routing strategy employs a combination of classification and recurrent neural networks (RNNs) for energy prediction and incorporates a shadow judgment method to improve accuracy. KEFSAR dynamically adjusts network routing based on solar intensity, improving both the energy efficiency and longevity of the network. Experimental results show that KEFSAR enhances prediction accuracy by 30%–50% and extends network lifetime by 10%–42%. Gökçen et al. [13] investigate the use of machine learning algorithms to forecast Li-ion battery discharge patterns in IoT devices. The article assumes that IoT devices are subject to random usage patterns. The authors evaluated various machine learning models, such as artificial neural networks (ANNs), Gaussian processes, and nonlinear regression, to predict battery capacity and internal resistance changes as a function of discharged energy. The discharge patterns are modeled using data from the NASA Ames prognostics data repository. The ANN model, with a radial basis function and a single hidden layer of 20 neurons, outperformed other models, achieving high prediction accuracy with |$R^{2}=1.0000$| and a low normalized mean square error. Wang et al. [14] introduce KVFL, a fuzzing framework tailored for IoT web servers. Unlike conventional fuzzing techniques, KVFL leverages key-value-based persistent fuzzing to effectively identify vulnerabilities in resource-constrained IoT devices. IoT web servers, widely used for management and data collection, are an attractive attack surface. KVFL operates by flipping specific data bits and injecting anomalous inputs to expose weaknesses. The framework enhances the fuzzing process by incorporating domain-specific key-value pairs that help achieve a high vulnerability detection rate while minimizing performance overhead. KVFL also supports continuous fuzzing sessions, increasing its potential for discovering complex bugs that require prolonged fuzzing periods. Experimental results show KVFL detects security flaws faster and with higher accuracy compared with traditional fuzzers. Cao et al. [15] propose an efficient deep learning-based approach to intrusion detection in IoT networks, addressing the increasing security risks in the IoT environment. The method leverages a convolutional neural network (CNN) to automatically extract features from network traffic data. A combination of CNN and an RNN is employed to enhance the detection of malicious activity by analyzing temporal patterns in the data. The approach was tested on the NSL-KDD dataset, achieving high accuracy and low false positive rates. The authors highlight that their system requires less manual feature engineering, improving efficiency compared with traditional methods. Moreover, the proposed model is scalable and can adapt to the high-volume, dynamic data characteristic of IoT systems. This makes the model well suited for real-time intrusion detection in large IoT infrastructures. Dung et al. [16] propose CAIMP, which is a cross-architecture malware detection and prediction framework for IoT devices. The proposed method relies on static features, such as opcode sequences, to detect malware across different hardware architectures. Using a machine learning model trained on these static features, CAIMP achieves high accuracy in detecting malware, reaching up to 99.4% in experiments. The approach is particularly effective in addressing the challenge of detecting malware that can operate on various hardware platforms, a key concern in heterogeneous IoT environments. CAIMP also includes predictive capabilities to anticipate malware behavior before full execution. The authors emphasize that CAIMP offers significant improvements in malware detection across IoT devices while maintaining a low false-positive rate. Zeng et al. [17] present a privacy-preserving framework for IoT data sharing using federated learning (FL) and generative adversarial networks (GANs). Federated learning enables decentralized data training on local devices, ensuring that sensitive data never leave the device. However, IoT data are vulnerable to privacy attacks. To mitigate this, the authors propose an enhanced FL model using GANs to generate synthetic data while protecting privacy. The model introduces noise to the GAN during training to prevent privacy leakage while preserving data utility. The proposed solution effectively balances data utility and privacy, as demonstrated in experimental results showing minimal performance degradation compared with traditional methods. Gómez et al. [18] propose a forensic methodology to investigate cyber incidents in IoT environments. The authors address the growing need for forensic tools capable of handling the complexity and scale of IoT systems. The methodology integrates traditional forensic models with IoT-specific adaptations, ensuring the ability to trace, collect, and analyze data effectively in distributed and heterogeneous networks. Key features include a layered approach to forensic investigation, addressing data volatility, and the need for secure evidence collection. Additionally, the method considers both technical and legal constraints, ensuring that the forensic process adheres to legal requirements while maintaining data integrity. Experimental validation demonstrates the methodology’s efficacy in identifying and addressing IoT-related cyber incidents. Zhang et al. [19] focus on the challenges of verifying the robustness of DNNs in smart IoT devices. As IoT devices are exposed to diverse environments and adversarial attacks, ensuring the robustness of their DNNs is critical to maintaining their functionality and security. The authors propose a formal verification method to improve the efficiency and accuracy of verifying DNN robustness in IoT systems. The key contribution is a tight linear approximation technique, which enables the efficient handling of complex neural network structures without sacrificing verification precision. This method helps in detecting adversarial inputs, reducing the risk of incorrect classifications in IoT devices. The results show that the proposed method outperforms previous approaches, especially in terms of scalability and computational efficiency. Noor [20] presents an innovative approach to crowd management using IoT technologies combined with behavior analysis. With growing populations, managing large crowds, especially during events like the Hajj pilgrimage, is crucial for safety. The proposed system uses IoT devices, such as cameras, to collect data and monitor crowds in real-time. A behavior analysis algorithm based on hidden Markov models is employed to identify normal and abnormal behaviors by analyzing spatio-temporal video segments. The approach allows for the classification of crowd dynamics and provides decision-makers with real-time insights to manage crowds effectively. Experimental results, based on real datasets from the Hajj pilgrimage, show that the system performs well in real-time environments. Antony and Singh [21] present a blockchain-based public key infrastructure designed for IoT-based healthcare systems. The proposed system leverages blockchain’s decentralized nature to overcome security challenges commonly faced in IoT networks. Using elliptic curve cryptography and smart contracts, the system allows secure communication and data exchange between IoT devices in healthcare environments. The article highlights how the integration of blockchain with IoT enhances data integrity, transparency, and resistance to cyberattacks. The system uses a secure key generation and distribution mechanism that ensures robustness against common threats, such as man-in-the-middle and replay attacks. Furthermore, the use of smart contracts ensures automated validation and revocation of certificates, reducing the need for a centralized authority. The proposed infrastructure addresses scalability and security issues in healthcare IoT applications, ensuring patient data privacy and secure communication across distributed IoT devices. Abid et al. [22] propose a blockchain and smart contract-based framework designed to address access control challenges in smart healthcare systems. Traditional access control models often rely on a centralized third party, which raises transparency and privacy concerns. To overcome this, the framework leverages blockchain technology and smart contracts to provide a decentralized, trustworthy solution. It integrates the generalized temporal role-based access control model to enforce fine-grained policies with temporal constraints. An initial implementation of the framework was evaluated for performance, showing linear increases in cost relative to policy complexity. Experimental results indicate that the approach is secure, efficient, and requires low operational costs. ChatGPT 4o has been used to summarize and extract keywords from articles.