Improvement And Application Of Parallel Distributed Algorithm Based On Raft Protocol

With the rapid growth of Internet data volume and concurrency,stand-alone systems have been difficult to meet the huge business needs.Internet companies have adopted distributed systems to expand the hardware resources owned by the system and reduce the risks caused by concurrent impacts.Multiple copies are an important means to solve the distributed single point problem and achieve load balancing.The consistency of data between multiple copies needs to be guaranteed by the consistency protocol.The famous distributed consistency protocol called Raft,is hard to synchronize data with high-efficiency.Raft's performance issues are more prominent in high-concurrency,large-scale cluster scenarios.Therefore,Raft Protocol based Parallel Distributed Algorithm is designed and implemented to improve the efficiency of distributed algorithms while ensuring distributed consistency.Although the Raft algorithm has good comprehensibility,the implementation of the traditional Raft algorithm has poor concurrency performance in high concurrency scenarios,and the election time is uncontrollable.On the basis of retaining the data characteristics that the Raft algorithm is easy to understand and easy to implement,the process of leader election and log synchronization is redesigned.An election algorithm based on synchronous clock is proposed.The election process is divided into three phases:initiating voting,collecting votes,and broadcasting results according to the time block to complete the election,so that the election is triggered at a specific physical time and completed within a fixed time.For the problem of log synchronization efficiency,a caching scheme was specifically designed to submit client requests in batches.In the implementation,increase the equipment gateway,decouple the consistency system and specific business functions,and reduce the load of the consistency system.After testing and experimentation,the Raft Protocol based Parallel Distributed Algorithm can correctly select the master in the state of no master,and meet the data consistency and availability.In the scenario of large-scale distributed clusters and high-end concurrent requests from clients,the above algorithms perform well in terms of election efficiency and log submission delay.