Research On Distributed Consensus Algorithm Of Low-Orbit Satellite Network Based On Raft Protocol

With the development of small satellite technology and inter-satellite communication technology and the rise of the concept of "Internet +" in recent years,satellite networking will become a trend in the future,and space-based Internet composed of different satellites may be realized in the future.In short,space-based Internet is a mobile distributed Internet composed of a group of satellites that communicate through inter-satellite links and cooperate with each other to complete tasks.As the space-based Internet is a distributed satellite network,information is copied and transmitted in the satellite network,which brings about the problem of distributed consensus.Consensus problem has a long research history in computer science,and it is an important foundation of distributed field.The existing distributed consensus algorithm of the ground Internet is mainly to keep the consensus of each replica in the process of data replication,which is only suitable for the network with high throughput and low latency.Due to the high-speed movement of satellites,the inter-satellite links is intermittent,the bandwidth is limited,and the communication delay between satellites is long.This does not meet the low latency requirements of the current consensus protocol,so the distributed consensus algorithm of the ground Internet is not fully applicable to the satellite network.Therefore,the extensive Raft consensus algorithm is selected for optimization and modification to adapt to the high dynamic and large delay environment of the satellite network.Raft consensus algorithm is used in satellite network environment,which has the following shortcomings: Raft election algorithm uses random time-out for election,due to the characteristics of high dynamic and large delay of satellite network,candidates may not receive enough votes after the end of election time-out,resulting in election failure and greatly increasing the time for electing leaders;Raft algorithm does not distinguish between reading,write operation,the leader writes any user request to the log by interacting with the followers,in the face of the high concurrency scenario of a large number of user requests,the user request cannot be processed in time;At the same time,the Raft algorithm writes the read operation that does not need to change the state value to the log,and does not use the log data of the followers,which will lead to the decline of the system read performance.In order to solve the problems of long time of leader election,log copy one by one,low efficiency of read operation in satellite network,raft consensus algorithm is optimized from three aspects: leader election algorithm,log copy and fast read,so as to design and implement distributed consensus algorithm in satellite network.A randomized bull algorithm is proposed,which can reduce leader recovery time and message delivery by directly selecting nodes with high priority.By adding cache in satellite nodes,the replication strategy of log entries can be changed from one by one to batch,which can effectively improve the efficiency of log replication.For the read operation that does not change the replica state,the vector clock mechanism is introduced to improve the read operation that requires leader and follower interaction messages in the raft algorithm to be processed in the follower node,which can meet the weak consistency and reduce the leader's load at the same time.The algorithm test is carried out in the satellite network simulation environment,using polar orbit satellite constellation architecture,with a total of 60 LEO satellites,randomly select 5 LEO satellites as service nodes,and carry out the consensus algorithm simulation test.The simulation results show that the randomized bull election algorithm can elect new leader in 2.3 seconds by reducing the number of election rounds,which improves the performance by 34% compared with the raft algorithm;with the increases the cache area,the performance of write operation throughput is doubled,and the average latency of users is slightly lower than that of raft algorithm;by adding fast read mechanism,the throughput performance of read request is improved by 143%,and the average latency of user read operation is reduced by 68%.