Motivation & Fit

In CnosDB concept design documentation (docs/reference/concept_design/compress.md), Gorilla and Pco are presented as the primary floating-point compression algorithms. While Gorilla is efficient for temporal XOR patterns and Pco provides high compression via tANS entropy coding, decimal-scaled time-series floats (such as sensor measurements, telemetry, market prices, and coordinates) benefit substantially from adaptive decimal extraction.

I would like to propose adding an introduction to ALP (Adaptive Lossless floating-Point Compression) and its pure Rust implementation fastalp (available on crates.io) to both Chinese and English concept documentation.

Summary of Changes

• docs/reference/concept_design/compress.md: Added ALP / fastalp section detailing its working principle (adaptive sampling, 10^e decimal scaling, Frame-of-Reference/Delta-ALP, vectorized bitpacking, and lossless exception handling) and applicability comparison with Gorilla and Pco.
• i18n/en/docusaurus-plugin-content-docs/current/reference/concept_design/compress.md: Added the corresponding English documentation.

Key Architectural Highlights

• Zero heap allocation: streaming compress_into and decompress_into interfaces.
• SIMD bitpacking: vectorized bit-unpacking routines delivering high-throughput scans.
• Adaptive dual-mode encoding: automatic selection between FOR and Delta-ALP for stationary vs monotonic series.
• Lossless guarantees: 100% bit-exact roundtrips with dedicated exception storage for non-decimal outliers.

Measured Results

On standard benchmarks and real-world time-series datasets:

Testing

• Verified markdown formatting and link integrity across Chinese and English documentation files.