An Improved Low Delay Distributed Consensus Algorithm in WAN

Abstract

In order to solve the problem that the number of Quorum nodes in two-phase of WAN distributed consensus algorithm is large and the location relationship between leader and client is ignored, a Paxos variant algorithm—Smallest Quorum Paxos(SQPaxos)—was proposed. In terms of quorums, SQPaxos determines the minimum number of nodes in the first phase Quorums and second phase Quorums, establishes the mapping relationship between proposal number and node number, and allocates nodes for each proposal to form Quorums, so as to meet the Quorum intersection condition and reduce consensus latency. In terms of the location relationship between leader and client, an adaptive leader allocation mechanism was proposed. Based on the multi-leader, by counting the frequency of requests from clients in different regions, the leader is adaptively transferred to the regions with high operation frequency to reduce the latency between leader and client. The experimental results show that SQPaxos which realizes the minimization of two-phase Quorums can greatly reduce the consensus latency. At the same time, the algorithm using adaptive leader allocation mechanism can reduce latency compared with the algorithm without using the mochanism or using other leader allocation mechanisms.

Keywords

• distributed consensus algorithm
• paxos
• quorum
• multi-paxos
• leader node