Design And Optimization Of Data Consistency Protocol For Distributed Storage System

With the rise of big data,the distributed storage technology has become increasingly important.Compared to centralized storage system,distributed storage system has remarkable advantages in cost,extensibility,availability and so forth.However,the architecture of distributed storage system also poses many challenges for data replication and synchronization.First,distributed storage system usually relies on asynchronous communications,and the data nodes in the system often break down due to unpredictable factors,making the distributed storage system difficult to achieve data consistency.Moreover,according to the theory of CAP,there exists data mutual restraint relationship among strong consistency,system availability,and partition fault tolerance.Therefore,it is nontrivial to ensure data consistency limiting to such complex environments and constraints.To answer these challenges above,this dissertation first implements a distributed data consistency module based on the well-known Paxos protocol,according to the characteristics of the distributed storage system.On this basis,we design a data consistency protocol that has advantages in simplified process,high availability and read/write synchronization.The main contents and results of this dissertation are summarized as follows:(1)We implement a Paxos protocol based distributed data consistency module.The module can receive data operation requests sent by multiple clients in parallel simultaneously,and generates the same data operation sequence acquired by each data node in the system.When each data node sequentially executes the data operation request in the sequence,the final consistency of the distributed storage system can be achieved.(2)We optimize the Paxos protocol by extending the effective period of the proposal epoch.The proposed scheme can effectively avoid the times of message exchange since Acceptor 's access is preempted by other Proposers,and therefore make the delay of the protocol reduced.(3)We further improve the proposed protocol by selecting some date node in as the Master node in the system.This scheme enables the system with read/write synchronization that outperforms the well-known Paxos protocol.In the end,we launch a set of experiments to verify the validity,effectiveness and efficiency by virtue of the proposed protocol.