The Internet of Things (IoT) and Edge-Cloud Computing have been trending technologies over the past few years. In this work, we introduce the Enhanced Optimized-Greedy Nominator Heuristic (EO-GNH), a framework designed to optimize machine learning (ML) and artificial intelligence (AI) application placement in edge environments, aiming to improve Quality of Service (QoS). Developed specifically for sectors such as smart agriculture, industry, and healthcare, EO-GNH integrates asynchronous MapReduce and parallel meta-heuristics to effectively manage AI applications, focusing on execution performance, resource utilization, and infrastructure resilience. The framework carefully addresses the distribution challenges of AI applications, especially Service Function Chains (SFCs), in edge-cloud infrastructures. It contains Data Flow Management, which covers aspects of data storage and data privacy, and also considers factors like regional adaptations, mobile access, and AI model refinement. EO-GNH ensures high availability for forecasting, prediction, and training AI models, operating efficiently within a geo-distributed infrastructure. The proposed strategies within EO-GNH emphasize concurrent multi-node execution, enhancing AI application placement by improving execution time, dependability, and cost-effectiveness. The efficiency of EO-GNH is demonstrated through its impact on QoS in real-time resource management across three application domains, highlighting its adaptability and potential in diverse cross-domain IoT-based environments. • EO-GNH optimizes AI application placement in edge computing environments for improved QoS performance • EO-GNH leverages Parsl for parallel meta-heuristic execution and optimized workload deployment • EO-GNH is faster than distributed NSGA-II in finding solutions to multi-objective optimization problems • EO-GNH delivers edge AI capabilities from federated learning to real-time inference • EO-GNH transforms IoT operations across healthcare, industry, and agriculture applications

With the rise of digital infrastructure and Internet of Things (IoT), a substantial amount of data is continuously generated that needs to be processed efficiently. While modern artificial intelligence (AI) approaches have shown good capabilities in handling large volumes of data, their excessive demands for memory and processing power result in very high utilization of resources. In this work, we propose ToSiM-IoT, an optimization framework that introduces a layer selection approach to identify an ideal mix of active, and inactive layers, using a genetic algorithm for model training. Next, we design a pruning mechanism that identifies performance-critical features using heatmap visualization, during model inference, and eliminates the remaining features. Two machine learning (ML) models: 1) InceptionV3 and 2) VGG16, have been evaluated on an agricultural weed detection scenario, using the DeepWeeds image classification dataset. Experimental results demonstrate that our framework can achieve a significant reduction in model size and training time, while maintaining high accuracy, for both models. Therefore, this approach provides the potential to be efficiently deployed on intelligent IoT systems where computational capabilities are limited.

Abstract The quality of machine learning (ML) models deployed in dynamic environments tends to decline over time due to disparities between the data used for training and the upcoming data available for prediction, which is commonly known as drift. Therefore, it is important for ML models to be capable of detecting any changes or drift in the data distribution and updating the ML model accordingly. This study presents various drift detection techniques to identify drift in the survival outcomes of patients who underwent cardiac surgery. Additionally, this study proposes several drift adaptation strategies, such as adaptive learning, incremental learning, and ensemble learning. Through a detailed analysis of the results, the study confirms the superior performance of ensemble model, achieving a minimum mean absolute error (MAE) of 10.684 and 2.827 for predicting hospital stay and ICU stay, respectively. Furthermore, the models that incorporate a drift adaptive framework exhibit superior performance compared to the models that do not include such a framework.

Deep Neural Networks (DNNs) have gained widespread popularity in different domain applications due to their dominant performance. Despite the prevalence of massively parallel multi-core processor architectures, adopting large DNN models in embedded systems remains challenging, as most embedded applications are designed with single-core processors in mind. This limits DNN adoption in embedded systems due to inefficient leveraging of model parallelization and workload partitioning. Prior solutions attempt to address these challenges using data and model parallelism. However, they lack in finding optimal DNN model partitions and distributing them efficiently to achieve improved performance. This paper proposes a DNN model parallelism framework to accelerate model training by finding the optimal number of model partitions and resource provisions. The proposed framework combines data and model parallelism techniques to optimize the parallel processing of DNNs for embedded applications. In addition, it implements the pipeline execution of the partitioned models and integrates a task controller to manage the computing resources. The experimental results for image object detection demonstrate the applicability of our proposed framework in estimating the latest execution time and reducing overall model training time by almost 44.87% compared to the baseline AlexNet convolutional neural network (CNN) model.

Abstract The Internet of Things (IoT) has gained widespread importance in recent time. However, the related issues of security and privacy persist in such IoT networks. Owing to device limitations in terms of computational power and storage, standard protection approaches cannot be deployed. In this article, we propose a lightweight distributed intrusion detection system (IDS) framework, called FCAFE‐BNET (Fog based Context Aware Feature Extraction using BranchyNET). The proposed FCAFE‐BNET approach considers versatile network conditions, such as varying bandwidths and data loads, while allocating inference tasks to cloud/edge resources. FCAFE‐BNET is able to adjust to dynamic network conditions. This can be advantageous for applications with particular quality of service requirements, such as video streaming or real‐time communication, ensuring a steady and reliable performance. Early exit deep neural networks (DNNs) have been employed for faster inference generation at the edge. Often, the weights that the model learns in the initial layer may be sufficiently qualified to perform the required classification tasks. Instead of using subsequent layers of DNNs for generating the inference, we have employed the early‐exit mechanism in the DNNs. Such DNNs help to predict a wide range of testing samples through these early‐exit branches, upon crossing a threshold. This method maintains the confidence values corresponding to the inference. Employing this approach, we achieved a faster inference, with significantly high accuracy. Comparative studies exploit manual feature extraction techniques, that can potentially overlook certain valuable patterns, thus degrading classification performance. The proposed framework converts textual/tabular data into 2‐D images, allowing the DNN model to autonomously learns its own features. This conversion scheme facilitated the identification of various intrusion types, ranging from 5 to 14 different categories. FCAFE‐BNET works for both network‐based and host‐based IDS: NIDS and HIDS. Our experiments demonstrate that, in comparison with recent approaches, FCAFE‐BNET achieves a 39.12%–50.23% reduction in the total inference time on benchmark real‐world datasets, such as: NSL‐KDD, UNSW‐NB 15, ToN_IoT, and ADFA_LD.

Serverless computing offers opportunities for auto-scaling, a pay-for-use cost model, quicker deployment and faster updates to support computing services. Apache OpenWhisk is one such open-source, distributed serverless platform that can be used to execute user functions in a stateless manner. We conduct a performance analysis of OpenWhisk on an edge-cloud continuum, using a function chain of video analysis applications. We consider a combination of Raspberry Pi and cloud nodes to deploy OpenWhisk, modifying a number of parameters, such as maximum memory limit and runtime, to investigate application behaviours. The five main factors considered are: cold and warm activation, memory and input size, CPU architecture, runtime packages used, and concurrent invocations. The results have been evaluated using initialization, and execution time, minimum memory requirement, inference time and accuracy.

With the increasing demand for edge computing in cyber-physical system (CPS) applications, ensuring the safety and reliability of machine learning models running on edge devices during online model training and inference is essential. Although data and model parallelism offer significant advantages for large machine learning model training, adopting parallel computing architecture in edge networks is challenging. It introduces safety concerns while splitting and integrating machine learning models over different computing nodes, which can pose risks to the integrity and reliability of the system. Therefore, online model training and inference in edge networks require a safe parallel computing architecture to achieve improved performance with optimal resource utilization. To address this challenge, we propose an efficient machine learning model partitioning algorithm that considers the safety constraint and requirements of edge networks and includes the triple-modular redundancy (TMR) technique for trusted computation. Our proposed approach achieves a significant speedup of approximately 56.3% in net training time compared to the non-partitioning approach, making it more efficient and suitable for real-time applications in edge networks.

Summary Fog computing offers cloud‐like facilities at the network edge, delivering reduced response times to latency sensitive applications. It comprises of fog devices/micro data centers/cloudlets located between users and the cloud data center. Fog devices are generally susceptible to privacy, security, and trust issues. We propose RT‐TADS (Real Time‐Trust Aware Dynamic Scheduling), a scheduling algorithm that accounts for privacy, trust and real‐time performance. To compute the trustworthiness of fog devices, we propose a trust computation model. This model factors in direct and recommended trust techniques for each fog device, and updates their aggregated trust values at regular intervals. User tasks are tagged as: private, semi‐private, and public. Fog devices are classified as: extremely highly trusted, highly trusted, normal trusted, low trusted, and untrusted. RT‐TADS maps the input jobs according to their privacy constraints on trustworthy fog devices, which increases the overall Success Ratio, hence improving real‐time performance. Using the Bitbrain dataset, the real‐time performance of RT‐TADS has been demonstrated, versus comparable algorithms. The results indicate that the proposed RT‐TADS offers an average improvement of 13%, 45%, and 71% in task success ratio compared to RLTCM , no‐trust , and cdc‐only respectively.

The unprecedented growth of Internet of Things (IoT) is leading to its increased usage in various domains, such as manufacturing, health, and smart cities. A majority of IoT applications are autonomic, i.e., they operate under minimal/no human intervention, and make decisions/actuations based on machine-to-machine communication and data analytics. A key challenge in the development of such applications is the ability to measure their quality while they are working in a diverse and heterogeneous IoT ecosystem. In this article, we propose an agent-based IoT-Quality Watch (IoT-QWatch) framework that provides the ability to measure IoT quality metrics at each stage of the autonomic IoT application life cycle running in the IoT ecosystem. We envision that IoT-QWatch will enable the development of a new generation of quality-aware autonomic IoT applications that are able to be resilient to the heterogeneous and uncertain nature of IoT ecosystems. We present architectural details and implementation of IoT-QWatch, and corresponding models used to measure IoT quality metrics at different stages. We conduct extensive experiments using a real-world IoT test bed from the domain of manufacturing to validate the efficacy of IoT-QWatch. Experimental outcomes provide promising results in realizing IoT-QWatch in real-world deployment, while the framework itself offers significant extensibility to include new models for measuring IoT quality metrics.

Cloud computing is widely used to support offloaded data processing for various applications. However, latency constrained data processing has requirements that may not always be suitable for cloud-based processing. Fog computing brings processing closer to data generation sources, by reducing propagation and data transfer delays. It is a viable alternative for processing tasks with real-time requirements. We propose a scheduling algorithm$RTH^{2}S$(RealTimeHeterogeneousHierarchicalScheduling) for a set of real-time tasks on a heterogeneous integrated fog-cloud architecture. We consider a hierarchical model for fog nodes, with nodes at higher tiers having greater computational capacity than nodes at lower tiers, though with greater latency from data generation sources. Tasks with various profiles have been considered. For the regular profile jobs, we use least laxity first (LLF) to find the preferred fog node for scheduling. In case of “tagged” profiles, based on their tag values, the jobs are split in order to finish execution before the deadline, or the LLF heuristic is used. Using HPC2N workload traces across 3.5 years of activity, the real-time performance of$RTH^{2}S$versus comparable algorithms is demonstrated. We also consider Microsoft Azure-based costs for the proposed algorithm. Our proposed approach is validated using both simulation (to demonstrate scale up) as well as a lab-based testbed.

Abstract Fog computing has emerged as a viable framework for processing delay sensitive applications. Modern applications consist of latency‐sensitive and latency‐tolerant jobs, leading to fog architectures that are often multi‐tiered/hierarchical. FiFSA (hierarchical first fog scheduling algorithm) and EFSA (hierarchical elected fog scheduling algorithm) are capable of scheduling both online and batch jobs on hierarchical fog‐cloud architectures. We consider heterogeneity in computing capacity—both among fog devices in separate layers, and among fog devices in the same layers. In general, online jobs with modest cpu requirements are scheduled on lower tier fog devices, and batch jobs with significant cpu requirements are scheduled on higher tier‐fog nodes, or the cloud data center (cdc). FiFSA assigns jobs to the first fog device with sufficient spare capacity. EFSA employs a MinMin heuristic that assigns jobs to the fog device that results in minimum completion time, while considering fog load. The performance of the proposed algorithms has been evaluated on a real‐life workload trace, using both simulation scenarios and a prototype testbed. FiFSA and EFSA offer an improvement of 19% to 70% in completion times and an improvement of 42% to 72% in system cost over other comparable algorithms.

Due to the significant communication delay to user tasks, the cloud is not ideal for executing real-time tasks with stringent deadlines. Fog computing consists of low computation capability fog nodes, or cloudlets located in proximity to the source of the data generation: the users. These cloudlets are ideal for executing tasks that have early deadlines. In this paper, we propose algorithms that schedule a set of real-time tasks on such an embedded-fog-cloud architecture. We consider hard, firm and soft tasks. The execution framework consists of embedded, fog and cloud processors. Tasks are scheduled on appropriate processors based on their deadline requirements. In general, hard real-time tasks are executed on embedded processors, firm real-time tasks on fog processors, and soft real-time tasks on cloud processors. We also propose a sufficient schedulability condition. Simulation results from the CERIT trace as well as test-bed results show that the proposed algorithms offer superior performance as compared to algorithms that do not employ fog processors. Employing an$Embedded-fog-cloud$architecture offers an improvement of 62.37 percent for real-time Success Ratio (SR) and 35 percent for Average Response Time as compared to scheduling tasks on the$cloud$alone.

Fog computing is a promising option for time sensitive vehicular over-the-air (OTA) updates, as it can offer enhanced network durability and lower communication delays, as compared to the cloud. Fog node utilization for updates is non-deterministic, largely owing to the patterns in vehicular traffic. The resultant over provisioning of resources manifests itself in increased communication and handover delays. Based on an analysis of the regional traffic pattern for a particular time period, our proposed algorithm determines the optimal number of fog nodes required for OTA updates. In order to pinpoint the traffic load and perform fog node distribution, we employ k-means clustering. The efficacy of our proposed approach is demonstrated using a case study that considers handover delay, propagation delay, transmission rate and vehicular mobility to predict the OTA update time. We employ a machine learning model for predicting the communication delay between fog devices and vehicles. Using the European WiFi hotspot signal strength NYC dataset and the 5G dataset, we observe that the proposed approach increases the net reserve fog resources by 26.57 percent on an average, and reduces the OTA update time by 5.34 percent. We test the scalability of the proposed approach by analyzing the performance in terms of average throughput while varying the number of vehicles and OTA update size. We observe that a system with less traffic and small update size overall delivers a higher average throughput of 46 Mbps versus one with more traffic and large update size overall, which provides an average throughput of 30 Mbps. The performance of the proposed OTA update scheme on simulations has been corroborated by implementation on a real-world testbed.

Due to advancements in technology, electrical appliances are now inter-connected. The goal of Internet -of-things (IoT) is to access every appliance or device through the Internet. This is done in order to operate these gadgets from remote locations. The goal is to improve our day-to-day life. However, this technology raises serious privacy and security issues. As IoT devices are resource-constrained, it is impractical to secure them using traditional approaches. Hence, a light-weight Intrusion detection system (IDS) is required. In this work, we implement a machine learning based Network Intrusion Detection (NID) system in a multi-node fog environment using a Raspberry Pi cluster on a local area network. The proposed Pi-IDS system has been evaluated on ADFA-LD datasets. These datasets comprise of new generation system calls for various attacks on different applications. The proposed fog architecture offers significant advantages in terms of latency, energy consumption and cost over traditional cloud or dedicated personal computer systems. The experiments show that we are able to achieve a Recall of 89%in ADFA-LD with the XGBoost model. The proposed system was able to predict intrusion with an inference time 130 ms in comparison to Cloud with 735 ms, with an estimated running cost of 201 INR/month in comparison to the Cloud cost of 2051 INR/month.

Fog computing extends the capability of cloud services to support latency sensitive applications. Adding fog computing nodes in proximity to a data generation/ actuation source can support data analysis tasks that have stringent deadline constraints. We introduce a real time, security-aware scheduling algorithm that can execute over a fog environment [1 , 2] . The applications we consider comprise of: (i) interactive applications which are less compute intensive, but require faster response time; (ii) computationally intensive batch applications which can tolerate some delay in execution. From a security perspective, applications are divided into three categories: public, private and semi-private which must be hosted over trusted, semi-trusted and untrusted resources. We propose the architecture and implementation of a distributed orchestrator for fog computing, able to combine task requirements (both performance and security) and resource properties.

Current IoT applications generate huge volumes of complex data that requires agile analysis in order to obtain deep insights, often by applying Machine Learning (ML) techniques. Support vector machine (SVM) is one such ML technique that has been used in object detection, image classification, text categorization and Pattern Recognition. However, training even a simple SVM model on big data takes a significant amount of computational time. Due to this, the model is unable to react and adapt in real-time. There is an urgent need to speedup the training process. Since organizations typically use the cloud for this data processing, accelerating the training process has the advantage of bringing down costs. In this paper, we propose a model partitioning approach that partitions the tasks of Stochastic Gradient Descent based Support Vector Machines (SGD-SVM) on various edge devices for concurrent computation, thus reducing the training time significantly. The proposed partitioning mechanism not only brings down the training time but also maintains the approximate accuracy over the centralized cloud approach. With a goal of developing a smart objection detection system, we conduct experiments to evaluate the performance of the proposed method using SGD-SVM on an edge based architecture. The results illustrate that the proposed approach significantly reduces the training time by 47%, while decreasing the accuracy by 2%, and offering an optimal number of partitions.

Fog computing brings the cloud closer to a user with the help of a micro data center ($mdc$), leading to lower response times for delay sensitive applications.RT-SANE(Real-TimeSecurityAware scheduling on theNetworkEdge) supports batch and interactive applications, taking account of their deadline and security constraints. RT-SANE chooses between an$mdc$(in proximity to a user) and a cloud data center ($cdc$) by taking account of network delay and security tags. Jobs submitted by a user are tagged as: private, semi-private and public, and$mdcs$and$cdcs$are classified as: trusted, semi-trusted and untrusted. RT-SANE executes private jobs on a user’s local$mdcs$or pre-trusted$cdcs$, and semi-private and public jobs on remote$mdcs$and$cdcs$. A security and performance-aware distributed orchestration architecture and protocol is made use of in RT-SANE. For evaluation, workload traces from the CERIT-SC Cloud system are used. The effect of slow executing straggler jobs on the Fog framework are also considered, involving migration of such jobs. Experiments reveal thatRT-SANEoffers a higher “success ratio” (successfully completed jobs) to comparable algorithms, including consideration of security tags.

Despite reduced network latency and resilience, fog computing has not been leveraged for vehicular Over-the-Air (OTA) updates. Due to vehicle mobility and traffic, the resource utilization of fog nodes is almost non-deterministic, which increases the delay in communication and handover. In this paper, we propose an approach for distributing fog nodes by analyzing the vehicular traffic pattern in a region. The proposed method: (a) finds the optimal number of fog nodes for a specific time interval based on the traffic pattern of a region and (b) maximizes the net reserve resources enabling specific fog nodes. To do so, we use the k-means algorithm to identify traffic load and distribute the fog nodes using our proposed algorithm to maximize fog resource utilization. We present a case study of OTA updates that considers vehicle mobility, data transmission rate, propagation delay and handover delay to predict the required update time. The experimental results demonstrate that the proposed method of fog node allocation extends the net reserve resources by 30.92% on an average, and reduces the OTA update time.

Summary Fog networks have attracted the attention of researchers recently. The idea is that a part of the computation of a job/application can be performed by fog devices that are located at the network edge, close to the users. Executing latency sensitive applications on the cloud may not be feasible, owing to the significant communication delay involved between the user and the cloud data center (cdc). By the time the application traverses the network and reaches the cloud data center, it might already be too late. However, fog devices, also known as mobile data centers (mdcs), are capable of executing such latency sensitive applications. In this paper, we study the problem of balancing the application load while taking account of security constraints of jobs, across various mdcs in a fog network. In case a particular mdc does not have sufficient capacity to execute a job, the job needs to be migrated to some other mdc. To this end, we propose three heuristic algorithms: minimum distance, minimum load, and minimum hop distance and load (MHDL). In addition, we also propose an ILP‐based algorithm called load balancing aware scheduling ILP (LASILP) for solving the task mapping and scheduling problem. The performance of the proposed algorithms have been compared with the cloud only algorithm and another heuristic algorithm called fog‐cloud‐placement (FCP). Simulation results performed on real‐life workload traces reveal that the MHDL heuristic performs better as compared to other scheduling policies in the fog computing environment while meeting application privacy requirements.

Duplication and dynamic voltage/frequency scaling (DVFS) creates an interesting trade-off for scheduling task graphs on multiprocessors to improve energy consumption and schedule length (or makespan). With DVFS, tasks are made to run on low voltages, which decreases their computation power. However, it also increases their execution costs and hence, may increase the schedule length. Furthermore, applying DVFS on processors does not impact the communication delay/energy consumption. Duplicating a task on multiple processors reduces the communication delay among them, which further reduces the schedule length. Although duplication reduces the communication energy among processors, it also increases the overall computation energy. In this paper, we explore this trade-off between duplication and DVFS, and propose a polynomial time heuristic to schedule task graphs on heterogeneous multiprocessors. The tasks are carefully duplicated with DVFS to reduce its impact on the computation energy. The results demonstrate that the proposed algorithm is able to effectively balance the makespan and energy consumption over other algorithms in various scenarios.

Energy efficiency along with enhanced performance are two important goals of scheduling on multiprocessors. This paper proposes a Contention-aware, Energy Efficient, Duplication based Mixed Integer Programming (CEEDMIP) formulation for scheduling task graphs on heterogeneous multiprocessors, interconnected in a distributed system or a network on chip architecture. The effect of duplication is studied with respect to minimizing: the makespan, the total energy for processing tasks and messages on processors and network resources respectively and the tardiness of tasks with respect to their deadlines. Optimizing the use of duplication with MIP provides both energy efficiency and performance by reducing the communication energy consumption and the communication latency. The contention awareness gives a more accurate estimation of the energy consumption. We also propose a corner case that allows the scheduling of a parent task copy after a copy of the child task which may lead to efficient schedules. It has been observed that the proposed MIP with a clustering based heuristic provides scalability and gives 10-30 percent improvement in energy with improved makespan and accuracy when compared with other duplication based energy aware algorithms.

This paper proposes a Real-Time Duplication-Based Algorithm (RT-DBA) for scheduling precedence-related periodic tasks with hard deadlines on networks of workstations (NOWs). We have utilized selective subtask duplication that enables some tasks to have earlier start times, which enables additional tasks (and, hence, task sets) to finish before their deadlines, thereby increasing the schedulability of a real-time application. We strongly believe that duplication can be used as a tool for obtaining a better quality of service (QoS) from the real-time heterogeneous system, and this is our major contribution. We have taken both the computation and the communication heterogeneities into account while modeling such a system. Both data and control dependencies between the tasks have also been considered. Our algorithm exhibits scalability, fully exploits the underlying parallelism, and is capable of scheduling an application, even if the available number of processors is less than the required number of processors. Based on extensive simulation studies, we observe that RT-DBA offers an enhanced success ratio as compared to other scheduling schemes when communication is a dominant factor.