Deep Neural Networks (DNNs) are increasingly adopted across various industries, driving the demand for deploying their capabilities on mobile devices. However, current mobile inference frameworks often rely on a single processor to execute each model inference, limiting hardware utilization and leading to suboptimal performance and energy efficiency. Expanding DNN accessibility on mobile platforms requires more adaptive and resource-efficient solutions to meet increasing computational demands without compromising device functionality. Nevertheless, performing parallel inference of multiple DNNs on heterogeneous processors remains a significant challenge. Existing studies have explored partitioning DNN operations into subgraphs to enable parallel execution across heterogeneous processors. However, these approaches typically generate excessive subgraphs based solely on hardware compatibility, increasing scheduling complexity and memory management overhead. To address these limitations, we propose the Advanced Multi-DNN Model Scheduling (ADMS) strategy that optimizes multi-DNN inference across heterogeneous processors on mobile devices. ADMS constructs an offline subgraph partitioning strategy that considers both hardware support for operations and scheduling granularity. It also employs a processor-state-aware scheduling algorithm to dynamically balance workloads based on real-time system conditions. This ensures efficient workload distribution and maximizes the utilization of available processors. Experimental results demonstrate that, compared to vanilla inference frameworks, ADMS achieves a 4.04× reduction in multi-DNN inference latency.

Sharing Internet of Things (IoT) data across different sectors, such as in smart cities, becomes complex due to heterogeneity. This poses challenges related to a lack of interoperability, data quality issues and lack of context information, and a lack of data veracity (or accuracy). In addition, there are privacy concerns as IoT data may contain personally identifiable information. To address the above challenges, this paper presents a novel semantic technology-based framework that enables data sharing in a GDPR-compliant manner while ensuring that the data shared is interoperable, contains required context information, is of acceptable quality, and is accurate and trustworthy. The proposed framework also accounts for the edge/fog, an upcoming computing paradigm for the IoT to support real-time decisions. We evaluate the performance of the proposed framework with two different edge and fog-edge scenarios using resource-constrained IoT devices, such as the Raspberry Pi. In addition, we also evaluate shared data quality, interoperability and veracity. Our key finding is that the proposed framework can be employed on IoT devices with limited resources due to its low CPU and memory utilization for analytics operations and data transformation and migration operations. The low overhead of the framework supports real-time decision making. In addition, the 100% accuracy of our evaluation of the data quality and veracity based on 180 different observations demonstrates that the proposed framework can guarantee both data quality and veracity.

The applications that are deployed in the cloud to provide services to the users encompass a large number of interconnected dependent cloud components. Multiple identical components are scheduled to run concurrently in order to handle unexpected failures and provide uninterrupted service to the end user, which introduces resource overhead problem for the cloud service provider. Furthermore such resource-intensive fault tolerant strategies bring extra monetary overhead to the cloud service provider and eventually to the cloud users. In order to address these issues, a novel fault tolerant strategy based on the significance level of each component is developed. The communication topology among the application components, their historical performance, failure rate, failure impact on other components, dependencies among them, etc., are used to rank those application components to further decide on the importance of one component over others. Based on the rank, a Markov Decision Process (MDP) model is presented to determine the number of replicas that varies from one component to another. A rigorous performance evaluation is carried out using some of the most common practically useful metrics such as, recovery time upon a fault, average number of components needed, number of parallel components successfully executed, etc., to quote a few, with similar component ranking and fault tolerant strategies. Simulation results demonstrate that the proposed algorithm reduces the required number of virtual and physical machines by approximately 10% and 4.2%, respectively, compared to other similar algorithms.

In Cloud Computing, the tenants opting for the Infrastructure as a Service (IaaS) send the resource requirements to the Cloud Service Provider (CSP) in the form of Virtual Network (VN) consisting of a set of inter-connected Virtual Machines (VM). Embedding the VN onto the existing physical network is known as Virtual Network Embedding (VNE) problem. One of the major research challenges is to allocate the physical resources such that the failure of the physical resources would bring less impact onto the users’ service. Additionally, the major challenge is to handle the embedding process of growing number of incoming users’ VNs from the algorithm design point-of-view. Considering both of the above-mentioned research issues, a novel Failure aware Semi-Centralized VNE (FSC-VNE) algorithm is proposed for the Fat-Tree data center network with the goal to reduce the impact of the resource failure onto the existing users. The impact of failure of the Physical Machines (PMs), physical links and network devices are taken into account while allocating the resources to the users. The beauty of the proposed algorithm is that the VMs are assigned to different PMs in a semi-centralized manner. In other words, the embedding algorithm is executed by multiple physical servers in order to concurrently embed the VMs of a VN and reduces the embedding time. Extensive simulation results show that the proposed algorithm can outperform over other VNE algorithms.

Thousands of high-end physical servers are used to fulfill the huge resource demand of diverse applications or services, ranging from healthcare data analytic services to gaming services. The network latency, as one of the major limitations of cloud computing, becomes the primary reason for introducing fog computing by pushing the computing environment towards the edge of the network. The ability to offer computing environments in close proximity to the user's device improves the delivery of high-quality services. The majority of the research is devoted to providing the high quality of services using either fog or cloud environment. In this paper, a novel deep reinforcement learning-based service dispersal approach for fog and cloud computing (DRLSD-FC) is adopted for offering the service using both environments simultaneously. The request to avail services is sliced and dispersed between the nearby fog and cloud environments. By taking advantage of cloud resources, the proposed approach minimizes the workload on the fog environment without compromising the service quality. The proposed approach is implemented using the Keras framework. Implementation results show that DRLSD-FC can outperform over other related approaches.

The amount of data generated by millions of connected IoT sensors and devices is growing exponentially. The need to extract relevant information from this data in modern and future generation computing system, necessitates efficient data handling and processing platforms that can migrate such big data from one location to other locations seamlessly and securely, and can provide a way to preprocess and analyze that data before migrating to the final destination. Various data pipeline architectures have been proposed allowing the data administrator/user to handle the data migration operation efficiently. However, the modern data pipeline architectures do not offer built-in functionalities for ensuring data veracity, which includes data accuracy, trustworthiness and security. Furthermore, allowing the intermediate data to be processed, especially in the serverless computing environment, is becoming a cumbersome task. In order to fill this research gap, this paper introduces an efficient and novel data pipeline architecture, named as CCoDaMiC (Coherent Coordination of Data Migration and Computation), which brings both the data migration operation and its computation together into one place. This also ensures that the data delivered to the next destination/pipeline block is accurate and secure. The proposed framework is implemented in private OpenStack environment and Apache Nifi.

The recent advances in virtualization technology have enabled the sharing of computing and networking resources of cloud data centers among multiple users. Virtual Network Embedding (VNE) is highly important and is an integral part of the cloud resource management. The lack of historical knowledge on cloud functioning and inability to foresee the future resource demand are two fundamental shortcomings of the traditional VNE approaches. The consequence of those shortcomings is the inefficient embedding of virtual resources on Substrate Nodes (SNs). On the contrary, application of Artificial Intelligence (AI) in VNE is still in the premature stage and needs further investigation. Considering the underlying complexity of VNE that includes numerous parameters, intelligent solutions are required to utilize the cloud resources efficiently via careful selection of appropriate SNs for the VNE. In this paper, Reinforcement Learning based prediction model is designed for the efficient Multi-stage Virtual Network Embedding (MUVINE) among the cloud data centers. The proposed MUVINE scheme is extensively simulated and evaluated against the recent state-of-the-art schemes. The simulation outcomes show that the proposed MUVINE scheme consistently outperforms over the existing schemes and provides the promising results.

The ability to fulfil the resource demand in runtime is encouraging the businesses to migrate to cloud. Recently, to provide real-time cloud services and to save network resources, fog computing is introduced. To further improve the quality of service in delivery process, Artificial Intelligence is being applied extensively. However, the state-of-the-art in this regard is still immature as it mainly focuses at either fog or cloud. To address this issue, a novel reinforcement learning-based personalized service delivery (RLPSD) mechanism is proposed in this paper, which allows the service provider to combine the fog and cloud environments, while providing the service. RLPSD distributes the user's service requests between fog and cloud, considering the users' constraints (e.g. the distance from fog), thus resulting in personalized service delivery. The proposed RLPSD algorithm is implemented and evaluated in terms of its success rate, percentage of service requests' distribution, learning rate, discount factor, etc.

Virtual network embedding (VNE) is the process of embedding the set of interconnected virtual machines onto the set of interconnected physical servers (PSs) in the cloud computing environment. The level of complexity of VNE problem increases when a large number of virtual machines with a set of resource demand need to be embedded onto a network of thousands of PSs. The key challenge of VNE is the efficient mapping of virtual networks (VNs), which may have dynamic resource demands. Existing solutions mainly emphasize on the embedding of static VN resulting in poor resource utilization and very low acceptance rate. To tackle such level of complexity in VNE, a fitness-based dynamic virtual network embedding (DYVINE) algorithm is proposed with the goal to maximize the resource utilization by maximizing the acceptance rate. Local and global fitness values of the virtual machines and VN, respectively, are used to utilize the maximum amount of physical resources. The proposed VNE algorithm allows the VN to be dynamic, which indicates that the structure and resource demand can be changed during its execution time. Furthermore, in order to reduce the embedding time in each time slot, a set of PSs is selected to host the VN instead of considering thousands of PSs, which may significantly increase the embedding time. The proposed embedding mechanism is evaluated through extensive simulation and is compared with similar existing embedding algorithms, which outperforms over others.

Virtualization technology boosts up traditional computing concept to cloud computing by introducing Virtual Machines (VMs) over the Physical Machines (PMs), which enables the cloud service providers to share the limited computing and network resources among multiple users. Virtual resource mapping can be defined as the process of embedding multiple VMs and their network resource demand onto multiple inter-connected PMs. The existing mechanisms of resource mapping need to be efficient enough to minimize the number of PMs without compromising the deadline of the tasks assigned to the VMs, which is NP-hard. To deal with this problem, a Link based Virtual Resource Management (LVRM) algorithm is designed to map the VMs onto PMs based on the available and required resources of the PMs and VMs, respectively. The designed algorithm exploits the fact that the demanded network bandwidth among VMs should be given higher priority while allocating the physical resources to the inter-connected virtual machines as insufficient network bandwidth may detain the task execution. The proposed algorithm is evaluated by a discrete event simulator and is compared with similar virtual network embedded algorithms. Simulation results show that LVRM can outperform over other network embedded algorithms.