Design And Implementation Of Load Balancing Of Edge Server Cluster For Wearable Applications

With the development of a smart society and the increasing needs of people,intelligence has penetrated into all sectors of society and people's daily lives.Edge devices have spread to all aspects of society,such as smart homes and self-driving cars in the transportation field,and smart production robots in the smart manufacturing field,resulting in a significant increase in the number of devices connected to the Internet.In the face of such a large number of devices accessing and transmitting data,it has been difficult for cloud platforms to meet the network latency requirements of edge devices due to distance and bandwidth limitations.The rise of edge computing technology allows cloud computing resources such as storage and computation to be placed on the edge side close to the edge devices,and the edge server clusters can use their abundant resources to provide application services to the edge devices,effectively reducing the latency of edge devices accessing the network.In this paper,we first analyze the edge wearable scenario,summarize the characteristics of wearable application fixity and low latency,and clarify the necessity of using edge server clusters.By comparing the two most popular edge computing platforms,we analyze their advantages and disadvantages,determine the selection of the edge computing platform,and analyze the load balancing strategy of the existing edge computing platform K3 S,indicate the advantages and disadvantages of the current load balancing strategy,and give the necessity of designing a load balancing strategy for the actual edge scenario by combining the characteristics of wearable applications.A dynamic weighted polling load balancing strategy is designed based on the existing load balancing strategy,which integrates the detailed load and network fluctuations of back-end service nodes and provides fine-grained scheduling for different applications,analyzes the resource requirements of different applications through linear regression,and predicts the response time and processing time of back-end service nodes using exponentially weighted moving average.Then the edge computing platform is built based on the proposed dynamic weighted polling load balancing strategy,the overall structure of the edge platform is analyzed,and the platform is divided into master nodes and service nodes for detailed design.The master node is designed with information collection module,load balancing module and database module for load balancing scheduling of external requests,while the service node is designed with information collection module,data persistence module,container management module and mirror management module for processing requests scheduled by the master node.Finally,based on the face detection application,target detection application and database read/write application in the edge scenario,the performance of the platform is tested under low and high load scenarios of the cluster.The test results show that the load balancing strategy can effectively reduce the response time and failure rate of requests and improve the throughput of the cluster for the wearable application in this paper in terms of throughput,response time and request failure rate.