Designing Binary Encodings for JSON and VARIANT
Summary
The post argues that binary encodings of JSON can dramatically speed up repeated lookups by avoiding parsing, introduces a minimal binary node layout with type tags, length, and direct offsets for arrays/objects, and contrasts this with BSON/CBOR/MsgPack while tying to Parquet VARIANT; it includes benchmarks showing up to 2,346x speedup and discusses design trade-offs for workload-specific binary encodings.