Research On Optimization Of Master Election Algorithm For Distributed Data Exchange Platform

With the advent of the big data era,the amount of data generated by all walks of life has increased dramatically,the limitations of the traditional single-node server began to show.To solve the high concurrent problems,the distributed server has gradually become a better solution.In this context,in order to deal with the growing amount of ETC data,Wuhan ETC trading system has introduced a distributed server that is aim of re-building a high concurrent,high integrated data exchange platform.As the data exchange platform in order to be compatible with the past ETC single-database model used,it produced some consistency problems,which resulted in failure to meet the business data consistency.Because such problems have a certain degree of universality,based on the consistency of "multi-server single data source" in ETC project,this paper focuses on the main node election scheme in several commonly used distributed consistency algorithms.This paper focuses on several common distributed consistency protocols and analyzes them.After analysis,the authors found that the Raft algorithm's random timeout election of the master node mechanism is well suited to solve the problem of distributed server generation in single data source mode.At the same time,the authors found that Raft algorithm itself cannot be combined with real-time election server load conditions,long-running normal cluster performance in general.Therefore,this paper presents a load-based voting mechanism,which takes the CPU usage,memory usage,network bandwidth occupancy and disk usage as the evaluation criteria of the server load.The author introduces the load calculation into the new master node election mechanism by monitoring the real-time load data of the ETC server.At the same time,the communication election protocol and flow in the Raft algorithm are improved.Finally,the paper compares the performance of the Raft algorithm before and after the improvement.The results show that the election time in the ordinary case does not change much,but the server load gap is relatively large,the cluster access efficiency increased by at best 25%.To some extent,it enhances the performance of the system under abnormal circumstances,and affects the normal performance of the system slightly.