Research On Key Problems Of Task Deployment In Cloud-Edge Joint Datacenter Environment

With the rapid development and widespread application of Io T and 5G technologies,the growth of smart mobile devices and applications is exploding,network traffic and service data volume are also increasing rapidly,and users' demand for data processing speed and quality service experience is also increasing.Although cloud computing can provide computing and storage services for data generated by smart mobile devices to break through the limitation of limited resources of terminal devices,however,due to the long link between the cloud data center and users,and in the process of data transmission by factors such as network bandwidth,resulting in a larger service delay.The rise of edge computing provides a reliable technical support for Io T,which is closer to the source of data.It places latency-sensitive applications or tasks to the edge nodes for computing without going through longer links to upload to the cloud data center,which not only enables fast response for critical tasks but also effectively relieves the computational pressure on the cloud data center.At present,the cloud-edge joint computing paradigm has gradually replaced the traditional computing paradigm and provided new momentum for the new Internet infrastructure.Based on the collaboration between the cloud data center on the center side and the edge computing center on the edge side in terms of resources,networks,applications and other aspects,the computing and processing at the edge release the pressure of the cloud data center to a certain extent,and a more powerful cloud-edge joint data center has formed.The rapid development of the Internet industry and information technology in recent years has led to an increasing expansion of the data volume on the Internet at an alarming rate,and the quality of service and resource utilization efficiency of the joint cloud edge system has received focused attention.The task deployment method of the cloud edge joint system directly determines the user service quality and system resource utilization efficiency,therefore,the adoption of reasonable and efficient deployment strategies and methods to achieve the highest quality service capability and resource utilization efficiency of the system is the key concern of many researchers.This paper carries out research and exploration of the problem of efficient processing of tasks in the joint cloud-edge computing environment from the following three aspects: firstly,in order to efficiently process latency-sensitive tasks,the virtualized parallel computing capability provided by the edge computing platform with fast processing speed and low latency characteristics needs to be used to process computing tasks,and the cloud computing model is used as an aid at this stage to upload tasks with high resource demand to the cloud computing center for processing,thus improving the external service performance of the edge computing center to a certain extent.Secondly,based on the edge joint cloud model,the cloud computing platform is combined with many edge nodes to form a cloud-edge joint data center,aiming to achieve efficient processing of tasks and load balancing among cloud-edge computing nodes.Thirdly,based on the existing "cloud-edge-end" architecture,we explore the optimal processing nodes for large-scale lightweight tasks in the mobile edge computing environment to minimize the cost of system bandwidth resources and node energy consumption.In this paper,the key problems of efficient task deployment in the cloud-edge joint data center environment are studied.Firstly,the efficient task deployment method of joint cloud computing model in edge computing environment is studied.Subsequently,the efficient task deployment method in cloud-edge joint computing environment is studied.Finally,the efficient processing method of large-scale lightweight tasks in mobile edge computing environment is studied.The main contributions of this paper are as follows:(1)The efficient task deployment of the joint cloud computing model in edge computing environment is studied.Due to the limited computing resources of edge computing center and the overload of most hosts,the edge computing ability is reduced.To solve this problem,this paper proposes a deployment method HEELS(Heuristic Task Clustering Method and Glowworm Swarm Optimization Algorithm)based on a clustering heuristic task analysis and glowworm swarm optimization algorithm.First of all,the tasks with high resource requirement are filtered out by clustering method,and the clustering results are uploaded to the cloud computing center for deployment and computation by task offloading technology.Then,the glowworm swarm optimization algorithm is used in the edge computing center,and by combining the mathematical model of SCA algorithm into the optimization of step size,the glowworm swarm optimization algorithm is made to change from a fixed step size to an adaptive step size,thus having better global search capability in the early stage and better local convergence capability in the later stage.The experimental results show that the HEELS method effectively improves the success rate of deployment tasks and the performance of external services,while achieving load balancing between the nodes in the edge computing center and the cloud computing center.(2)The efficient task deployment in cloud-edge joint computing environment is studied in this paper.At present,the existing research usually uses unilateral computing center(cloud or edge)to implement task processing and computing,which may be limited by the performance of unilateral data center and the efficiency of resource utilization.To solve this problem,this paper proposes an efficient task deployment method JCETD(Joint Cloud and Edge Task Deployment)in a cloud-edge joint computing environment.Its main idea includes two parts: firstly,the idea of pruning algorithm is used to preprocess the cloud edge joint host set,remove the unreasonable physical host in the set,and get a non-dominated joint host set,and the complexity of the following algorithm is reduced effectively.Then,the task deployment problem is simulated as a process of deep reinforcement learning in the cloud-edge joint computing environment,through the exploration and utilization of the system environment,deploy tasks rationally and efficiently on the physical host of the cloud edge joint data center.The experimental results show that the proposed JCETD method significantly reduces the total completion time and the average response time,effectively optimizes the external service performance,and realizes the load balancing of the cloud edge joint computing system.(3)The efficient processing of large-scale lightweight tasks in mobile edge computing environment is studied.In the existing research,when the traditional methods and frameworks are used for lightweight task processing of mobile terminal,the high cost of bandwidth resources and high energy consumption caused by communication between nodes have become the key challenges.In order to cope with this challenge,this paper proposes a task deployment method ADIC(Task Allocation by Deep Learning and Incremental Learning in Mobile Edge Computing Environment)based on deep learning and incremental learning in mobile edge computing environment to minimize the overall network bandwidth cost and energy consumption between task processing nodes in wireless mesh networks.Firstly,we use the idea of deep feature learning to learn the features of each computing node and task in the mobile edge environment and construct the feature matrix as the input of the deep network model.Then,the input feature matrix is gradually analyzed through the constructed convolution layer and the maximum pooling layer,so as to obtain the wireless network backbone with probability tag corresponding to each processing node in the wireless mesh network.After getting the current deep network model,in the practical application process,based on the idea of incremental learning,we regularly update the deep model with new data samples that are continuously generated,so that the model has the better task deployment capability.The experimental results show that the proposed ADIC method minimizes the cost of bandwidth and the energy consumption of nodes in the parallel task processing of wireless networks.