| As the financial information technology plays more important role in financial business, it is an urgent task to build a financial matching system with high performance. Due to the high need for throughput, response latency, high availability and scalability, how to design and implement in terms of architecture as well as the functionality and coordination of modules is a big challenge.The dissertation discusses the drawbacks of the matching system based on database, and points out distributed and in-memory architecture is more applicable. Then it generalizes the distributed technology and concurrent programming models. After analyzing the needs and logics of financial business and key technical points, it proposes a high concurrent foreign exchange matching system with high performance. Then it conducts detailed analysis and in-depth study on architecture, pipelined processing of orders, asynchronized database persistence and failover mechanism. And based on the analysis and testing of symmetrical architecture, master-slave architecture has better performance under the scenario, and it simplifies and optimizes the data synchronization mechanism between master and slave nodes; for pipelined operation of orders, it implements a lock-free circular queue supporting high currency, and it makes it possible to execute the orders processing of different stages in parallel; the asynchronized database persistence guarantees the data persistence and reduces the I/O cost with the application of append-only log files; the failover mechanism provides failure detection and recovery with heartbeat packets and cluster management so that it ensures the high reliability of the system.According to the experiment test, the foreign exchange matching system proposed in this dissertation overwhelms the reference model with symmetrical architecture and a self-defined concurrent queue. The average throughput is 1.4 times, the latency reduces 25%, and it has better scalability and robustness. |
PDF Full Text Request