Research And Improvement Of Blockchain Consensus Mechanism

Appeared as the crucial technology of bitcoin,Blockchain has become a research hotspot in various fields in recent years.As a typical decentralized distributed data storage system,the consensus reached by all participants in a blockchain system requires a consensus mechanism to achieve it.In order to make the blockchain suitable for different application scenarios,there are already many different consensus mechanisms.With the further development of blockchain applications,more and more studies involving consensus mechanism.However,some existing consensus mechanisms still have some problems.In this paper,the classic distributed consensus mechanism in blockchain is analyzed and summarized firstly.The research is focus on the existing problems of PBFT consensus mechanism and its related improvement schemes.Based on this,the paper proposed two different improvements.Firstly,focusing on the problem that PBFT lacks dynamics,Voting-based Practical Byzantine Fault Tolerance(VPBFT)is proposed.A voting mechanism was introduced in VPBFT,and all nodes in VPBFT are divided into 4 types with different responsibilities which are voting nodes,ordinary nodes,candidate nodes,and production nodes.There is a certain quantitative relationship between different types of nodes.When the number of nodes changes,it could calculate based on the quantity relationship dynamiclly.Besides,the “two-phase” consensus model is used to instead of the “threephase” of PBFT that could reduce the consumption of network bandwidth.Through performance simulation analysis and compared with PBFT and its improved solutions,VPBFT is better than PBFT and its improved solutions in energy consumption,fault tolerance and delay.And VPBFT has dynamics,reliability and security.Secondly,focusing on the problem that PBFT lacks dynamics,and the large number of communications due to the increase in nodes,Machine learning-based Practical Byzantine Fault Tolerance,MLPBFT(MLPBFT)is proposed.MLPBFT uses the K-means clustering algorithm to complete the initial division of nodes.After consensus,the K-means clustering algorithm is automatically adjusted.When the number of nodes changes,all nodes would be divided into 4 types including malicious nodes and 3 types of nodes that perform consensus processes by the adjusted Kmeans clustering algorithm automatically.And the consensus process uses a "two-stage" model,which reduces consensus time and communication times.In addition,during the time waiting for the consensus result,the client node makes full use of its own resources to run a logistic regression algorithm to predict the consensus result,and temporarily stores data blocks in advance according to the predicted probability.After experimental testing and simulation analysis,MLPBFT is better than PBFT in terms of consensus time consumption,energy consumption,throughput,fault tolerance,and communication times while being dynamic and secure.