Availability Optimization Of Devices And Data In Internet Of Things

With the rapid development of Internet of Things(IoT)technology,IoT applications with high availability requirements such as smart industry,smart home,smart city,and autonomous driving are constantly emerging.IoT is also considered to be the third wave after the computer and Internet technology in the development process of the world information industry.In the era of the Internet of Everything,the massive and diverse IoT devices and data bring multiple challenges to IoT development,high system availability,and ultimately the quality of service(Qo S)improvement of IoT systems.On the one hand,as the basic platform for IoT systems to provide various services,the real-time and continuous work leads to the higher power consumption of IoT devices.This causes the temperature of IoT devices hardware to rise,and in turn degrades the lifespan of IoT devices.Moreover,the probability of IoT devices becoming unavailable due to various failures(such as transient failures and permanent failures,etc.)is continuously rising.On the other hand,due to the fragility of IoT devices,the instability of energy supply,and the unreliability of transmission,there are often low available IoT data with missing or outliers.Insufficient IoT data availability is a crucial reason for IoT systems to produce unexpected intermediate results and ultimately serious failures.Therefore,improving the availability of IoT devices and data is critical to enhancing the functionality and availability of IoT systems.However,the current availability optimization schemes for IoT devices and data have problems such as time-consuming simulation,high communication overhead,ignoring the characteristics of correlation and distribution of data,which lead to degradation of optimization.Therefore,to address the deficiencies of the existing solutions,our paper is dedicated to designing the availability optimization schemes for IoT devices and data.To be specific,we describe the contributions of this paper as follows.1.This paper studies the availability optimization problem of IoT devices integrating CPU and GPU under the temperature constraint.For independent real-time tasks running on such IoT devices,an artificial neural network(ANN)is first trained to quickly and accurately estimate the transient failure rate and temperature of the IoT system,which overcomes the time-consuming shortcomings of existing simulation-based schemes.Then,a device availability optimization scheme based on a feedback control is proposed,which judiciously determines the numbers of tasks and corresponding replicas in the IoT system at each time slot by iteratively feeding back the utilization and other resources of IoT systems under the constraint of temperature.Extensive simulation experiments are conducted in this paper.We compare the proposed method with multiple benchmarking methods in terms of the accuracy of trained artificial neural networks,the reliability of IoT systems,and the availability of IoT devices to measure the performance of the proposed IoT device availability optimization method.2.This paper studies the availability optimization problem of structured IoT data with correlation.We solve the problem of how to optimize the completeness and accuracy of IoT structured data under the premise of these data are correlated.This paper first proposes a K nearest neighbor(KNN)-based IoT structured data availability optimization scheme,which models the correlation between structured IoT data and finds an initial solution for data availability optimization based on the KNN algorithm.Then,an availability optimization scheme for IoT structured data using the orthogonal matching pursuit(OMP)method is presented,which makes the estimation of IoT data more accurate by iteratively revising the initial solution from the KNN-based availability optimization method.Finally,inspired by singular value decomposition(SVD),this paper proposes a uniqueness analysis scheme to verify the uniqueness and validity of the found final solution.This paper conducts extensive experiments on multiple real IoT structured datasets.Compared with various benchmarking methods in terms of completeness,accuracy,impact on the original data,and the time cost of the algorithm,the performance of the proposed IoT structured data availability optimization method is validated.3.This paper studies the availability optimization mechanism of IoT unstructured data using generative adversarial network(GAN),and is dedicated to solving the availability optimization problem of IoT unstructured data represented by images.This paper first proposes an efficient multi-discriminator conditional generative adversarial network(MDCGAN)architecture,which structurally consists of a generator and multiple discriminators.The proposed MDCGAN architecture can achieve data availability optimization without aggregating distributed IoT unstructured data.Then this paper designs a hint mechanism and parameter exchange mechanism to improve the performance of the proposed IoT unstructured data availability optimization scheme based on MDCGAN.The hint mechanism provides auxiliary external information for the GAN-based architecture to speed up the convergence of the MDCGAM model.The parameter exchange mechanism takes into account the characteristics of distributed data,and can be used to reduce the probability of model overfitting.This paper conducts extensive experiments on multiple real unstructured datasets.Compared with a variety of benchmarking methods in terms of the completeness and accuracy of IoT unstructured data,the impact of the proposed algorithm on the original data,and the time cost of the algorithm,the performance of the proposed IoT unstructured data availability optimization scheme is verified.The availability optimization technologies of IoT devices and data proposed in this paper can better improve the mean time to failure of IoT devices and the completeness and accuracy of IoT data,and can help promote the deployment of IoT in various high-availability applications.