Design And Implementation Of High Availability Cache Cluster

With the continuous development of cloud computing,data-intensive applications require more powerful and more usable storage systems to support their performance requirements.Disk-based storage has high stability,and relational databases are widely used.Disks as storage media,causing great pressure on the database,the database not only need to deal with the read and write of the disk,but also need to manage a large amount of structured data,such as creating an index and maintaining the index Such operations,the database will often become the bottleneck of the entire back-end system,so in large-scale back-end systems often use cache to ease the pressure on the database.This article first studied the existing problems of Memcached.Although the memory-based storage method has the characteristics of fast response,it will lose data due to the opportunity of the storage service instance,and cannot restore the data in memory after the restart,and the traditional The deployment of Memcached clusters is often managed in a single-agent manner,which may cause performance bottlenecks.In addition,traditional Memcached storage service instances cannot communicate with each other.This causes backup of data to be managed by clients,and the related synchronization plug-ins need to be client-side.Extra development is therefore not widely used.Based on the original Memcached cluster,this paper presents a DMEM distributed caching system based on the analysis of the original Memcached cluster.There are two main systems proposed in this paper.One obvious innovation: 1.Design a highly available caching system based on the concept of separation of weights and separations.Separation refers to the decoupling of cache set management functions from the data storage management functions,and the maintenance of cluster status through the Controller component,through the node_agent node.Manage actual data storage nodes and reduce application complexity by caching communication agents.2.Implement management functions of the master and backup machines based on the open source repcached plug-in.Add the state machine through the node_agent node to implement the failover function.Cache automatic synchronization recovery through repcached plugin.In this paper,after the requirement analysis,the DMEM system is designed and developed in detail.Finally,this paper verifies that the DMEMsystem meets the functional and performance requirements.