Compares performance between JSON, Gzipped JSON, JSON-Schema, Google Protocol Buffers, and Apache Avro.

$ python benchmark.py

There are a few steps to make sure you're running your benchmark on the latest and greatest releases of Python and Google's Protocol Buffers:

1. Update Python to latest version. I prefer using homebrew:

$ brew install python
#OR
$ brew upgrade python

1. Install / Upgrade Google Protocol Buffer Compiler

$ brew install protobuf
#OR
$ brew upgrade protobuf

1. Install dependencies

pip install -r requirements.txt

1. Upgrade Benchmark Dependencies

pip install --upgrade avro
pip install --upgrade protobuf
pip install --upgrade jsonschema

1. Delete Existing Address Book Python Class for Reading/Writing Protobuf

rm -f pb_schemas/addressbook_pb2.py

1. Compile Proto Schema Into Python Class

SRC_DIR='schemas' && DST_DIR='schemas' && protoc -I=$SRC_DIR --python_out=$DST_DIR $SRC_DIR/addressbook.proto

1. Run Benchmarks

$ python benchmark.py

Here are my results, running Python 2.7.9, jsonschema 2.4.0, Protobuf 2.6.1, and Avro 1.7.7 on an Early 2013 15" Macbook Pro Retina:

Notes:

• JSON and Gzipped JSON do not do schema validation
• JSON-Schema, Avro, and Protocol Buffers do schema validation

Results:

Results (indexed to JSON):

These bench mark results are when json, protobuf, and avro write operations iterated over a set of N objects and transformed each object into the respective format (json, protobuf, avro) and then writing to disk. Read operation was not benchmarked. Executed against a dataset of 100K objects.

Json is the fastest in this test, but protobuf and avro both clock at under 1 ms per record (~.1ms).