ALP: Adaptive Lossless Floating-Point Encoding in Apache Parquet
Summary
Apache Parquet introduces ALP (Adaptive Lossless floating-Point) Encoding, a lightweight floating-point encoding that offers compression ratios close to zstd with faster decompression and random-access support, designed for FLOAT and DOUBLE types. It targets common decimal-like data such as monetary values, geographic coordinates, and scientific measurements, while avoiding broad-exponent data like vector embeddings; adoption is already appearing in Parquet ecosystem implementations and platform releases.