A port of python-serialization-benchmark to Rust, using serde_json as the serializer.
I attempted to reproduce the workings of the Python objects as closely as possible, and the output of both the Python and Rust serializers is identical.
Clone this repo, and run
cargo bench
The results from cargo bench will be in nanoseconds, rather than seconds as in the Python benchmark.
Running both the Python and Rust benchmarks on my local machine (a late-2013 MacBook Pro with a 2.6GHz quad-core Intel Core i7), produced the following:
Library Many Objects (seconds) One Object (seconds) Relative
--------------------- ------------------------ ---------------------- ----------
serde_json (Rust) 0.00407219 0.00218101 0.48
serpyco 0.00872946 0.00450587 1
Custom 0.010844 0.00496244 1.19426
lima 0.0138385 0.00684261 1.56257
Pickle 0.015667 0.016381 2.4214
serpy 0.0538666 0.0283134 6.20914
Strainer 0.0799131 0.0385959 8.95399
Toasted Marshmallow 0.104154 0.0554521 12.0591
Colander 0.280941 0.142262 31.9752
Avro 0.442745 0.217734 49.9027
Lollipop 0.4566 0.223092 51.3544
Marshmallow 0.609148 0.307631 69.2675
kim 0.799352 0.388839 89.7742
Django REST Framework 0.724293 0.543168 95.7634
Each framework is asked to serialize a list of 2 objects a 1000 times, and then 1 object a 1000 times. See the original Python benchmark for more details.