一种基于Paxos算法的证券交易系统内存复制方法研究

近年来随着高速网络技术的发展与高频交易需求的增加，提升交易速度成为电子商务交易提供者的重要关 切。当前交易系统通常采用基于共享存储的主备机复制方法来保证高可用性与数据持久性，但因其存在持久化的性 能瓶颈而无法进一步降低延迟。为此，提出一种基于Paxos算法的内存数据复制方法，即通过消息传递完成主备机复 制，以保证结点间数据的一致性，容忍可能发生的良性故障;并以证券交易系统场景为例对其进行分析。实验结果表 明，相比基于共享存储的主备机复制，该方法在万兆以太网环境下可将交易系统订单处理延迟由毫秒级降至百微秒 级，并在主机故障时正确地完成热备切换。

Research on a Paxos-based Approach for Memory Data Replication in Stock Trading System

As the development of high-speed network technology and rising demand for high-frequency trading, upgra- ding trading speed has become the main focus of e-commerce trading system providers in recent years. Primary-backup replication based on shared storage is a classical approach to ensure high availability and data durability of trading sys- tans at present, but it is difficult to further reduce the system latency due to its persistence bottleneck. ho solve this problem,a Paxos-based approach for memory data replication was proposed and illustrated in the stock trading context. hhe approach accomplishes the primary-backup replication through messaging, can ensure the strong consistency of data replicas and tolerate possible benign failures. Experimental results show that compared with the shared storage replica- tion approach,this approach reduces the order processing latency of stock trading system from several milliseconds to several hundred microseconds in 10 Gbit Ethernet, and achieves hot failover correctly in lure. case of the primary host fai