Search

With the proliferation of the Internet of Things and smart devices, there exists an urge to address the critical computation demands of end users for several real-time applications. Fog computing (FC) targets to address the computation requirements nearer to the end user in almost real-time. Mainly, the fog environments (FE) exhibit characteristics of resource constraint and wide heterogeneity. In addition, the complex fog environment enforces the fog resources (FR) deployment either within fixed or distributed over random geographical locations. In the literature, there exist several solutions for the optimization of FR allocation. However, the location-dependent FR allocation has not been addressed thoroughly. Hence, this paper proposes a novel hybrid clustering-based deep Q-network for location-based optimal resource allocation (HCDQN-ORA) model. Here, the first objective is to minimize the search operation for identifying suitable FR and to meet the quality of service (QoS) demand using an enhanced fog resource clustering (E-FRC) algorithm. The second objective aims to achieve optimal FR allocation among the set of location-dependent FRs using reinforcement learning techniques. Problem formulation has been modelled using Markov decision process. A deep Q-network algorithm with two variants, namely enhanced optimal resource allocation using deep Q-network (EORA-DQN) and enhanced optimal resource allocation using deep Q-network with experience replay (EORA-DQN-ER), has been incorporated. To analyse the statically located and randomly distributed FR environment, the experimental results exhibit that, in the case of clustering of FRs, E-FRC is more efficient than conventional clustering algorithms. Next, the location-dependent optimal FR allocation problem incorporates the performance analysis based on the convergence of loss function and the parametric analysis based on batch size, respectively. Results exhibit that EORA-DQN performs better only for statically allotted resources. However, the proposed EORA-DQN-ER outperforms both types of FR with a success rate of 89.6% of optimal allocation with the batch size of 200. [ABSTRACT FROM AUTHOR]

Federated learning is a distributed machine learning approach that enables a large number of edge/end devices to perform on-device training for a single machine learning model, without having to share their own raw data. We consider in this paper a federated learning scenario wherein the local training is carried out on IoT devices and the global aggregation is done at the level of an edge server. One essential challenge in this emerging approach is IoT devices selection (also called scheduling), i.e., how to select the IoT devices to participate in the distributed training process. The existing approaches suggest to base the scheduling decision on the resource characteristics of the devices to guarantee that the selected devices would have enough resources to carry out the training. In this work, we argue that trust should be an integral part of the decision-making process and therefore design a trust establishment mechanism between the edge server and IoT devices. The trust mechanism aims to detect those IoT devices that over-utilize or under-utilize their resources during the local training. Thereafter, we introduce DDQN-Trust, a double deep Q learning-based selection algorithm that takes into account the trust scores and energy levels of the IoT devices to make appropriate scheduling decisions. Finally, we integrate our solution into four federated learning aggregation approaches, namely, FedAvg, FedProx, FedShare and FedSGD. Experiments conducted using a real-world dataset show that our DDQN-Trust solution always achieves better performance compared to two main benchmarks: the DQN and random scheduling algorithms. The results also reveal that FedProx outperforms the competitor aggregation models in terms of accuracy when integrated into our DDQN-Trust solution. [ABSTRACT FROM AUTHOR]

Summary: The cloud computing systems, such as the Internet of Things (IoT), are usually introduced with a three‐layer architecture (IoT‐Fog‐Cloud) for the task offloading that is a solution to compensate for resource constraints in these systems. Offloading at the right location is the most significant challenge in this field. It is more appropriate to offload tasks to fog than to cloud based on power and performance metrics, but its resources are more limited than the resources of the cloud. This paper tries to optimize these factors in the fog by specifying the number of usable servers in the fog. For this purpose, we model a fog computing system using the queueing theory. Furthermore, binary search and reinforcement learning algorithms are proposed to determine the minimum number of servers with the lowest power consumption. We evaluate the cost of the fog in different scenarios. By solving the model, we find that the proposed dispatching policy is very flexible and outperformed the known policies by up to 31% and in no case is it worse than either of them, and the overall offloading cost increases when fog rejects tasks with a high probability. Our offloading method is more effective than running all fog servers simultaneously, based on simulation results. It is evident from the similarities between the simulation results and those derived from the analytical method that the model and results are valid. [ABSTRACT FROM AUTHOR]