Research On Client Load Balancing And Data Synchronization Optimization In Zookeeper Cluster

Due to the emergence of Moore's Law bottleneck,the distributed system architecture model has become the first choice for major enterprises.In order to cope with fault tolerance across the equipment room and balance the access performance of each area,deployment across the computer room becomes a necessary choice.In the scenario where the ZooKeeper cluster is deployed in a remote room,the network delay between the different equipment rooms is large and unstable.The following two problems will occur: First,the client session will experience large fluctuations in access performance.Second,the cluster data synchronization efficiency will decrease as the number of clusters increases.In the aspect of ZooKeeper client session,through the in-depth analysis and simulation experiments of the client session process,this paper designs and implements the distributed server address list configuration center to solve the server address list as "hard coded",designed and implemented based on Fork/Join framework server network delay sorting to sort the network delay of each server,design and implement the hardware resource monitoring module,and classify the load,and formulate the synchronization synchronization and timing synchronization strategy to reduce the synchronization pressure of the registration center.This paper integrates the above three designs and proposes that the shortest network delay priority method can be used to stabilize and improve client read performance under the non-high load of ZooKeeper server,and the client load balancing mechanism in Zookeeper is reconstructed.The optimized client session API.In the aspect of ZooKeeper cluster data synchronization,through the in-depth analysis and simulation experiments of cluster data synchronization process,this paper establishes a time-response model of node voting ability,and points out that the main factors affecting voting ability are network delay,node computing power and disk reading.speed.In order to evaluate the voting ability of nodes,this paper uses the combined weights judgment method based on Lagrange optimal multiplier method to construct the evaluation model,and designs the automatic voting node based on node voting ability to improve the efficiency of data synchronization,and gives the automatic Configure the indicator calculation and algorithm design of the voting node.In order to compare the performance of client load balancing and data synchronization in the ZooKeeper cluster before and after optimization,this paper uses a cross-regional cloud server to build a ZooKeeper cluster for performance testing.In terms of client load balancing,the test results show that the overall average response time of the optimized client session is reduced by 36.1% compared with that before optimization.The optimized Beijing client read request is reduced by 68.8% compared with that before optimization.Client read requests are reduced by 34.8% compared to before optimization.In terms of data synchronization,the test results show that the data synchronization efficiency in the automatic configuration voting node is improved by 12.1% compared with the optimal situation in the manual configuration voting node,and is 70% higher than the worst case in the manual configuration voting node.At the same time,automatic configuration of voting nodes is better than manual configuration of voting nodes,both in terms of flexibility and accuracy.