MV-Raft:一种新的基于多维向量的联盟链共识算法

MV-Raft:A New Consensus Algorithm for Alliance Chain Based on Multidimensional Vectors

现有联盟链的默认算法Raft对于节点崩溃采用的是随机的机制,未考虑不同节点权重可能导致达成共识时间较长,且导致领导者节点日志同步压力较大的问题。针对该问题,提出一种基于多维向量的Raft算法——MV-Raft。算法先根据影响达成共识效率的因素构成特征值向量,然后在分布式系统启动或者单个节点新加入系统时向全网进行广播上述特征值向量,通过计算特征值向量从而判断节点间的相似度,最后,节点内部缓存相似度最高的n个节点在选举和同步过程中优先对缓存节点进行操作。通过仿真实验表明,MV-Raft算法可以使整个过程的耗时减少16%,让单个领导者节点的压力最多减少73%。

The default algorithm of Raft for existing alliance chains uses a random mechanism for node crashes,without considering that different node weights may result in longer consensus times and greater log synchronization pressures on leader nodes.To address this issue,a Raft algorithm based on multidimensional vectors is proposed,called MV-Raft.The algorithm first constructs an eigenvalue vector according to factors that affect the efficiency of reaching a consensus,and then broadcasts the above eigenvalue vector to the whole network when the distributed system is started or a single node is newly added to the system.The similarity between nodes is calculated based on the eigenvalue vector,and the n nodes with the highest similarity are cached within each node for prioritized operation during the election and synchronization process.The simulation results show that MV-Raft algorithm can reduce the time consumption of the whole process by 16%,and reduce the pressure of a single leader node by 73%at most.