benchmark performance of popular serialization libraries
- json
- protocol buffer
- msgpack
To run the benchmarks,
make test
make runresults
python2 (Python 2.7.9)
| encode | decode | |
|---|---|---|
| json | 0.46 | 0.59 |
| pb | 0.16 | 0.20 |
| msgpack | 3.73 | 4.23 |
python3 (Python 3.4.2)
| encode | decode | |
|---|---|---|
| json | 0.45 | 0.35 |
| pb | 0.16 | 0.20 |
| msgpack | 2.92 | 3.00 |
pypy (PyPy 2.4.0 with GCC 4.9.2, Python 2.7.8)
| encode | decode | |
|---|---|---|
| json | 0.37 | 0.11 |
| pb | 0.70 | 0.31 |
| msgpack | 0.32 | 0.71 |
| umsgpack | 0.22 | 0.40 |
| msgpack-pypy | 0.42 | 0.30 |
Note: msgpack-pypy is in beta state. Other libraries are in stable state.
summary
For py2/py3, pb > json >> msgpack. For pypy, json > pb = msgpack.
Json is a good choice. It doesn’t have a big performance penalty.
TODO Why msgpack has so bad performance under cpython?