CAJ | 학술논문

针对现有Dynamo式云存储架构中,节点间全局状态表一致性维护算法收敛时间长、通信代价高的问题,文中提出了一种基于种子节点的Gossip(GossipBasedonSeednodes,GBS)算法。该算法采用分层思想,将节点分为种子节点层和普通节点层,通过分层,缩小了Gossip传播随机域,减少收敛时间；通过普通节点定期向随机的种子节点报告策略,解决了Gossip传播后期由于随机带来的大量重复交互问题。通过仿真实验,证明了GBS算法在降低收敛时间的同时,有效地降低了通信开销,提高了维护效率,具有很高的稳定性和可靠性。
침대현유Dynamo식운존저가구중,절점간전국상태표일치성유호산법수렴시간장、통신대개고적문제,문중제출료일충기우충자절점적Gossip(GossipBasedonSeednodes,GBS)산법。해산법채용분층사상,장절점분위충자절점층화보통절점층,통과분층,축소료Gossip전파수궤역,감소수렴시간；통과보통절점정기향수궤적충자절점보고책략,해결료Gossip전파후기유우수궤대래적대량중복교호문제。통과방진실험,증명료GBS산법재강저수렴시간적동시,유효지강저료통신개소,제고료유호효솔,구유흔고적은정성화가고성。
In existing Dynamo cloud storage architecture system,the consistency maintenance algorithm of the global state table between nodes needs a long convergence time and high communication cost. To solve this problem,propose a GBS ( Gossip Based on Seed nodes) algorithm. This algorithm uses hierarchical thinking to divide nodes into a seed node layer and ordinary node layer,which shrinks the Gos-sip spread random field and reduces the maintenance convergence time. Nodes use regularly report to random seed nodes on strategies to solve the random repeated interaction problems of the late of Gossip spread. The simulation experiments show that the GBS algorithm can effectively reduce the maintenance convergence time and communication cost of the state table,improve the maintenance efficiency,and have a high reliability and stability.