This article details Ctrip's motivation, design, implementation, performance evaluation, and future plans for a custom MySQL storage engine called CFL that efficiently stores time‑series data by leveraging MySQL's replication and a sequential write‑optimized file format.
Time‑series data, defined by a timestamp field, appears everywhere, and the article explains how to choose an appropriate time‑series database by comparing two schema models, their query patterns, performance trade‑offs, and why modern solutions like Elasticsearch, columnar stores, and Druid excel at real‑time massive aggregation.
This article examines the evolution and challenges of time‑series storage, compares file‑based, LSM‑tree, and B‑tree approaches, and proposes a vector‑based method to efficiently handle writes, reads, query semantics, dimensions, and aggregation for modern big‑data applications.
