KungfuDB: A Low Latency In-Memory Time Series Database for Quantitative Trading Systems

Xiaodong Li, Yan Zhou, Wenkai Liu, Yizhi Zhang, Keren Dong

Abstract: Modern quantitative trading systems rely on low-latency time series databases for critical tasks such as tick data writing, reading and persistence, which support facilities like trading signal analysis and strategy back-testing. However, existing databases, even top-ranked solutions on DB-Engine, lack specialized optimizations required by these financial applications. To address this need, Kungfu Origin Technology and Hohai University develop KungfuDB, a low latency in-memory financial time series database. KungfuDB uses innovative log and page management algorithms and zero-copy memory techniques to achieve ultra-fast database operations with minimal latency. Experiments against leading databases like InfluxDB and TimescaleDB show that KungfuDB delivers up to 60x performance improvements in writing and reading tick data. Notably, KungfuDB is open-sourced in Github and successfully deployed by leading securities and fund management firms in China.

External IDs:doi:10.1007/978-981-95-4158-4_35