Distributed Consensus Algorithm Based On Peer-to-Peer Partial Ordering Arbitrament Model

Distributed consensus algorithm is the infrastructure of modern distributed computing system and distributed service framework,responsible for ensuring the existence of the united internal state view between distributed members.The formalization description and mathematical proof of the distributed consensus is provided by typical Paxos algorithm,laying the foundation of distributed consensus theory,and the emergence of Paxos algorithm family and main commerce as well as open-sourcing distributed consensus algorithm derived subsequently makes an outstanding contribution for the reliability,performance and engineering practice of distributed consensus algorithm.However,there are Live Lock problems on the Paxos family derived algorithm based on P2P(Peer-to-Peer)model,and there are uneven load and single point performance bottlenecks on the commercial distributed consensus algorithm of which mainstream based on master-slave model caused by Leader nodes.For the purpose of further optimizing the performance bottlenecks of distributed consensus algorithm while avoiding the production of live lock problems,this thesis proposes a study of the new constructed distributed consensus algorithm on the basis of distributed consensus basis theory and the main tasks are as follows:(1)Propose the basic framework of distributed partial ordering arbitration distributed consensus algorithm to solve Live Lock problems existed in typical Paxos and P2P model distributed consensus algorithm as the representative of the same algorithm.(2)Aiming at uneven load,failure single point and single point performance bottlenecks of distributed consensus system of which mainstream based on master-slave,the method of dynamic generation and migration of load center is proposed;(3)Aiming at excessive wide valid scope constrained by mainstream distributed consensus algorithm legal set,the constraint of legal set is only valid in minority members by new member division,and the number of member nodes that participate in a single vote is reduced to effectively reduce the average delay of convergence while the validity and high ability of algorithm is ensured.(4)Achieve rapid detection state differences and implement data sync by recursion hash technology called log chain to ensure the validity of distributed consensus system;(5)The modeling is established for Y-Paxos distributed consensus algorithm proposed by the paper through Opnet discrete event simulation,compared with typical Raft algorithm,and the results show that the Y-Paxos distributed consensus algorithm proposed by the paper has flush performance and more clear framework,by eliminate leader node,failure single point and load bottlenecks compared with traditional distributed consensus algorithm.