MessagePack vs JSON vs Others

When to use each:

JSON:

• Public APIs (browser-friendly).
• Config files (Hjson, JSON5 for human).
• Logging.
• Default for most web work.

MessagePack:

• Internal RPCs where you want JSON-like flexibility but smaller.
• Caches that need compact serialization.
• Languages without good Protobuf support.
• Schemaless storage.

Protobuf:

• Service-to-service in big systems.
• Tight schema control + evolution.
• gRPC.
• Google + many enterprises.

CBOR:

• Similar to MsgPack but IETF standardized.
• Used in some IoT (CoAP).

BSON:

• MongoDB internal.
• Larger than MsgPack but typed.

FlatBuffers / Cap'n Proto:

• Zero-copy access to structured data.
• Game state, real-time graphics.

Avro:

• Hadoop / Kafka ecosystems.
• Schema embedded with data (write).

Performance benchmarks (rough, vary by platform):

• JSON: 1x baseline.
• MsgPack encode: 2-3x faster.
• MsgPack size: 30-50% smaller.
• Protobuf encode: 3-5x faster than JSON.
• Protobuf size: 60-80% smaller than JSON.

Picking:

• Public + browser: JSON.
• Internal + flexibility: MsgPack.
• Internal + strict + perf: Protobuf.
• IoT + standardized: CBOR.

For new internal services: Protobuf is usually the right choice. MsgPack when you need JSON's flexibility minus the verbosity.