Comparing JSON, BEVE, MessagePack, CBOR, and Protocol Buffers for C++ binary serialization performance.

• JSON, BEVE, and CBOR use Glaze
• MessagePack uses msgpack-c
• Protocol Buffers use zpp_bits

Higher means BEVE is faster by that factor. Format: Write/Read

CBOR uses RFC 8746 typed arrays via Glaze.

Both BEVE and CBOR (via Glaze) store contiguous arrays of trivially copyable types as raw memory blocks, enabling efficient memcpy operations. This results in throughput exceeding 60 GB/s for typed arrays.

MessagePack encodes each element individually with type tags, resulting in significant overhead for numeric data.

Protobuf (via zpp_bits) also uses memcpy for packed repeated fields, but has additional overhead from varint length prefixes, field tags, and struct conversion between native C++ types and protobuf-compatible types.

• Protobuf achieves the smallest size for complex objects (344 B) due to field tag encoding and varint integers
• BEVE and CBOR achieve identical sizes for typed arrays via raw memory storage
• MessagePack has ~12% overhead for typed arrays due to per-element type tags

Note: msgpack-c is expected to be in the same directory as this repository on the branch cpp_master.

git clone https://github.com/msgpack/msgpack-c.git --branch cpp_master
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make
./binary_perf

Results are written to results.md.