基于排队论的交易系统时延分析  

QUEUING THEORY-BASED LATENCY ANALYSIS ON TRADING SYSTEMS

作　　者：黄寅飞[1] 王泊[1] 白硕[1] 

机构地区：[1]上海证券交易所技术开发部,上海200120

出　　处：《计算机应用与软件》2016年第11期326-330,共5页Computer Applications and Software

基　　金：国家科技支撑计划项目(2012BAH13F04)

摘　　要：低延迟是证券交易系统从小型机迁移到PC服务器过程中的主要挑战,为此需要进行时延度量和分析。首先设计实现基于Redis和Python的时延数据采集监控框架。其次提出队列级联方法,分析分布式系统时延构成。最后通过参数试验和数据分析,找到令吞吐量和时延最优的配置方法,并分析解决可扩展性和毛刺问题。将分布式系统建模为级联队列,在此基础上通过量化分析、性能调优和流控保护等手段,令时延指标显著提升。所提出的队列级联方法可为企业级关键业务实时系统的设计和调优提供参考。Low latency is the major challenge of stock-trading system when to be moved from minicomputers to PC servers,with regard to this it requires the latency measurement and analysis. First we designed the Redis and Python-based latency data collecting and monitoring framework and implemented it. Then we proposed the cascading queues method to analyse the latency composition in distributed systems.Finally we found the configuration way enabling the optimisation of throughput and latency through parameters test and data analyses,and studied the solutions for extensibility and microburst problems. In this paper we model the distributed systems as cascading queues,on this basis,we make a significant improvement in latency index by the means of quantitative analysis,performance tuning and flow control protection,etc. The cascading queues method proposed in the paper can provide references for design and tuning of enterprise-level missioncritical real-time systems.

关 键 词：时延度量 监控框架 低延迟 证券交易系统 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]