Research And Optimization Of Distributed Consistency Algorithm For Wide Area Distributed Storage System

With the arrival of the era of big data and the explosive growth of data,and the ever-increasing user demand for data services availability,scalability,and fault tolerance,the value and importance of wide-area distributed storage technology is becoming increasingly prominent.Replication technology is the key to achieve fast response of wide-area distributed storage systems,and the performance of data consistency is an important indicator for evaluating replication technology.At present,the most effective way to solve data consistency problems is to use a distributed coherence protocol with a leader node structure such as the Multi-Paxos protocol.However,as more and more distributed storage systems move to multiple data center architectures across different geographical regions worldwide,wide-area distributed storage systems face difficulties such as high protocol latency and low throughput.Under the constraint of the CAP theorem,traditional distributed coherence protocols cannot meet user requirements for consistency and availability in a wide area network environment.Therefore,how to effectively improve the availability of distributed consistency protocols in a WAN environment is a challenging topic.Regarding the issue above,in order to improve system throughput and latency,this paper studies the optimization of distributed data consistency protocol in WAN environment by optimizing the non-leader node consistency protocol.The main research contents and innovations are summarized as follows:Firstly,the theoretical analysis of the protocol degradation in the case of command conflicts in the typical non-leader node distributed consistency protocol EPaxos protocol is carried out.By introducing a distributed global clock,an improved EPaxos protocol based on timestamp sorting,namely T-EPaxos protocol,is proposed.The protocol changes the protocol degradation process when the EPaxos protocol command conflict occurs to the order according to the timestamp of the command,thereby effectively reducing the number of message transmissions caused by protocol degradation and improving system performance like delay and throughput.On this basis,according to the characteristics of wide-area distributed storage system,a hybrid consistency scheme H-Paxos with different consistency protocols in data center and across data centers is proposed.Among them,in the single data center in the multi-data center architecture across regions,the S-Paxos protocol with the leader node is used,and between the data centers,the T-EPaxos non-leader node consistency protocol proposed in the foregoing is used.In order to further improve the response speed of real-time scenarios such as disaster warning and emergency response,according to CAP theorem,defined distributed consistency strength and system availability,and based on this,a distributedlocal consistency replication framework for wide-area distributed systems is proposed.Simulation results show,the local packet of the proposed distributed local consistency replication framework has better delay and throughput performance.It provides new ideas and breakthroughs for the research of wide-area distributed storage systems.