Research On Dynamic Feedback Load Balancing Strategy

With the advent of the Internet plus era,Internet technology provides convenient services for people's lives at all times.While providing services,Internet technology also faces great challenges.The access to a large number of concurrent services has caused tremendous pressure on a single server.Server clusters are used to solve server-side pressures.The most important one is how load-balancing technology allocates tasks properly and makes the performance of the entire cluster system the best.The existing load balancing algorithm does not consider the load status of each node in the server cluster during operation,including CPU utilization,network bandwidth,memory usage,and IO read/write speed,resulting in the existing load balancing algorithm throughput and clustering.The overall utilization rate needs to be improved.Therefore,this paper proposes a dynamic load balancing algorithm designed to improve the throughput and utilization of the cluster,including the following three aspects:1)The DWA algorithm is proposed for the problem that the existing load balancing algorithm does not consider the load status of cluster nodes.First,the node load evaluation parameters need to be selected,and the Http request classification is divided into release type requests and transaction type requests,and the server load of the two types of requests is analyzed to select appropriate evaluation parameters.Then,the server load status is quantified using weighted multiplication.Finally,the DWA algorithm is proposed to perform task scheduling based on the load status of the current server node,so as to achieve more reasonable scheduling of tasks and improve the throughput of the cluster.2)The cluster load balancing strategy is the key to improving the server cluster.The dynamic feedback load balancing strategy is designed based on the DWA algorithm.It mainly includes three aspects: the collection of load information,the determination of load status,and the implementation of load strategy.A centralized information collection method is used to periodically collect load information of nodes and use UDP multicast to transmit information.In the process of determining the load information,the weight of the server is used to evaluate the load capacity of the server,and the AHP method is used to calculate the weight to avoid the error caused by manually setting manually.During the execution phase of the load policy,the average load threshold of the CPU is introduced to determine whether the multi-core CPU reaches the maximum workload,which avoids using the CPU utilization to determine the error caused by the server workload.3)In order to verify the effectiveness of the algorithm,this chapter conducts specific experimental verification and validates the effectiveness of the verification algorithm through the cluster throughput rate and the average response time.First of all,how many concurrent requests defined in the current experimental environment is a large number of requests.Since the server cluster is for a large number of requests,the advantage of the cluster can be reflected only when the amount of requests is large.Therefore,a comparative experiment is designed for the response time between the single server and the cluster server.A comparison experiment of the DWA algorithm and the weighted minimum connection method is designed,and the proposed algorithm is evaluated by the throughput rate and response time.Experimental results show that the DWA algorithm is superior to the weighted minimum connection method in throughput and response time.