Kakao Restructures Distributed Web Cache Wcache Architecture… 3x Performance Boost for Identical Content Processing

Kakao has significantly restructured the storage architecture and lock mechanism of its distributed web cache Wcache, which handles traffic. The previous version stored metadata in a centralized SQLite DB, causing a sharp increase in DB load during block replacement and resulting in performance drops of up to 80%.

Storage Structure Change: Introduction of JBF

To overcome the limitations of a centralized DB, Kakao introduced the Journaling BigFile (JBF) structure, which stores metadata inside the BigFile. JBF includes block-level journals, space maps, and B-Trees, enabling recovery and formatting at the individual file level, thereby resolving performance degradation issues during cache flushing. Additionally, Bloom Filters were applied to reduce redundant JBF searches, improving response speeds by 2–5x compared to the previous version.

Feature and Lock Structure Improvements

The metadata structure was split into Meta and Info types to support flexible storage of Vary Objects, and content size limits were removed. Furthermore, to resolve bottlenecks caused by concentrated requests for identical content, read/write locks were separated, and disk I/O and verification logic were isolated into separate actors. These measures improved the performance of identical content request processing by approximately 3x compared to the previous version.