Paxos-Based Memory Data Replication in Stock Trading System

摘要

Primary-backup replication based on shared storage is a classical approach to ensuring high availability and data durability of stock trading systems, but it is difficult to further reduce the system latency due to persistence bottleneck. To solve this problem, a memory data replication approach based on Paxos algorithm is proposed, which accomplishes primary-backup replication through messaging, ensures the strong consistency of data replicas and tolerates possible benign failures. Experiment results have shown that this approach, compared with the shared storage approach, manages to greatly reduce the order processing latency of trading systems and achieve hot failover correctly in the case of server failure.