| In order to improve the performance of storage and network bandwidth in distributed environment,it is a developing trend to combine erasurity coding with consistency algorithms,such as RS-Paxos and CRaft.In a network composed of 2F+1nodes,RS-Paxos and CRaft cannot tolerate both the liveness of F nodes and the improvement of system performance.In other words,when F nodes fail,the system can not continue to run or switch to the original full copy replication.How to effectively combine the erasure code with the consistency algorithm and ensure the high performance and high availability of the system is the main content of this paper.This paper designs a consistency algorithm BBD-Raft for placing erasure coding packages based on balanced block design.It designs different distribution matrices according to the parameters of erasure codes and BBD,and then obtains real-time healthy nodes in the system through heartbeat.Finally,it combines and distributes different numbers of coding packages to different nodes according to the distribution matrix,so as to meet the requirement that the whole system can always store enough complete data for recovery.Experiments show that BBD-Raft algorithm can detect healthy nodes in the system and distribute data packets according to the number of nodes.The system performance is also better than CRaft while ensuring the failure of F nodes.On the other hand,this paper designs a highly available key-value storage system based on BBD-Raft.The system uses the instructions of the log storage client to store data in different system nodes,and reduces the amount of data stored by each node through coding.This article describes in detail the design process and algorithm implementation,as well as data persistence based on badger database.Finally,a virtual cluster is constructed,the number of running nodes is controlled by pre-generated test cases,and read and write operations are injected into the system to verify the correctness and high availability of the system. |
PDF Full Text Request