Skip to content

Instantly share code, notes, and snippets.

@justinfx
Last active October 16, 2022 13:28
Show Gist options
  • Save justinfx/3174062 to your computer and use it in GitHub Desktop.
Save justinfx/3174062 to your computer and use it in GitHub Desktop.
Speed test of common serializers on python 2.7.2 (pickle, cPickle, ujson, cjson, simplejson, json, yajl, msgpack)
"""
Dependencies:
pip install tabulate simplejson ujson yajl msgpack
"""
from timeit import timeit
from tabulate import tabulate
setup = '''d = {
'words': """
Lorem ipsum dolor sit amet, consectetur adipiscing
elit. Mauris adipiscing adipiscing placerat.
Vestibulum augue augue,
pellentesque quis sollicitudin id, adipiscing.
""",
'list': list(range(100)),
'dict': dict((str(i),'a') for i in range(100)),
'int': 100,
'float': 100.123456
}'''
setup_pickle = '%s ; import pickle ; src = pickle.dumps(d)' % setup
setup_pickle2 = '%s ; import pickle ; src = pickle.dumps(d, 2)' % setup
setup_pickle3 = '%s ; import pickle ; src = pickle.dumps(d, 3)' % setup
setup_pickle4 = '%s ; import pickle ; src = pickle.dumps(d, 4)' % setup
setup_json = '%s ; import json; src = json.dumps(d)' % setup
setup_msgpack = '%s ; src = msgpack.dumps(d)' % setup
tests = [
# (title, setup, enc_test, dec_test)
('pickle (ascii)', 'import pickle; %s' % setup_pickle, 'pickle.dumps(d, 0)', 'pickle.loads(src)'),
('pickle (binary-2)', 'import pickle; %s' % setup_pickle2, 'pickle.dumps(d, 2)', 'pickle.loads(src)'),
('pickle (binary-3)', 'import pickle; %s' % setup_pickle3, 'pickle.dumps(d, 3)', 'pickle.loads(src)'),
('pickle (binary-4)', 'import pickle; %s' % setup_pickle4, 'pickle.dumps(d, 4)', 'pickle.loads(src)'),
('json', 'import json; %s' % setup_json, 'json.dumps(d)', 'json.loads(src)'),
('simplejson-3.17.0', 'import simplejson; %s' % setup_json, 'simplejson.dumps(d)', 'simplejson.loads(src)'),
('ujson-1.35', 'import ujson; %s' % setup_json, 'ujson.dumps(d)', 'ujson.loads(src)'),
('yajl 0.3.5', 'import yajl; %s' % setup_json, 'yajl.dumps(d)', 'yajl.loads(src)'),
('msgpack-0.6.2', 'import msgpack; %s' % setup_msgpack, 'msgpack.dumps(d)', 'msgpack.loads(src)'),
]
loops = 15000
enc_table = []
dec_table = []
print("Running tests (%d loops each)" % loops)
for title, mod, enc, dec in tests:
print(title)
print(" [Encode]", enc)
result = timeit(enc, mod, number=loops)
enc_table.append([title, result])
print(" [Decode]", dec)
result = timeit(dec, mod, number=loops)
dec_table.append([title, result])
enc_table.sort(key=lambda x: x[1])
enc_table.insert(0, ['Package', 'Seconds'])
dec_table.sort(key=lambda x: x[1])
dec_table.insert(0, ['Package', 'Seconds'])
print("\nEncoding Test (%d loops)" % loops)
print(tabulate(enc_table, headers="firstrow"))
print("\nDecoding Test (%d loops)" % loops)
print(tabulate(dec_table, headers="firstrow"))
"""
Running tests (15000 loops each)
pickle (ascii)
[Encode] pickle.dumps(d, 0)
[Decode] pickle.loads(src)
pickle (binary-2)
[Encode] pickle.dumps(d, 2)
[Decode] pickle.loads(src)
pickle (binary-3)
[Encode] pickle.dumps(d, 3)
[Decode] pickle.loads(src)
pickle (binary-4)
[Encode] pickle.dumps(d, 4)
[Decode] pickle.loads(src)
json
[Encode] json.dumps(d)
[Decode] json.loads(src)
simplejson-3.17.0
[Encode] simplejson.dumps(d)
[Decode] simplejson.loads(src)
ujson-1.35
[Encode] ujson.dumps(d)
[Decode] ujson.loads(src)
yajl 0.3.5
[Encode] yajl.dumps(d)
[Decode] yajl.loads(src)
msgpack-0.6.2
[Encode] msgpack.dumps(d)
[Decode] msgpack.loads(src)
Encoding Test (15000 loops)
Package Seconds
----------------- ---------
pickle (binary-2) 0.141743
pickle (binary-3) 0.149022
pickle (binary-4) 0.151039
msgpack-0.6.2 0.155419
ujson-1.35 0.180287
yajl 0.3.5 0.388522
json 0.440383
simplejson-3.17.0 0.484677
pickle (ascii) 0.661336
Decoding Test (15000 loops)
Package Seconds
----------------- ---------
msgpack-0.6.2 0.136735
pickle (binary-4) 0.172758
pickle (ascii) 0.173402
pickle (binary-3) 0.175427
pickle (binary-2) 0.178137
ujson-1.35 0.206923
json 0.328846
yajl 0.3.5 0.438112
simplejson-3.17.0 0.475276
"""
@halsafar
Copy link

I would also like to know how msg pack suddenly went from the winner to worse than pythons native pickle. Python3 pickle must be heavily optimized.

@justinfx
Copy link
Author

I've updated the test and ran them again against python-3.7.5.
msgpack-python has been deprecated in favor of msgpack, which is now the fastest again. It uses cython bindings which means it should be faster than any pure python solution. You can also now see the difference between the different binary protocol levels of pickle.

Running tests (15000 loops each)
pickle (ascii)
  [Encode] pickle.dumps(d, 0)
  [Decode] pickle.loads(src)
pickle (binary-2)
  [Encode] pickle.dumps(d, 2)
  [Decode] pickle.loads(src)
pickle (binary-3)
  [Encode] pickle.dumps(d, 3)
  [Decode] pickle.loads(src)
pickle (binary-4)
  [Encode] pickle.dumps(d, 4)
  [Decode] pickle.loads(src)
json
  [Encode] json.dumps(d)
  [Decode] json.loads(src)
simplejson-3.17.0
  [Encode] simplejson.dumps(d)
  [Decode] simplejson.loads(src)
ujson-1.35
  [Encode] ujson.dumps(d)
  [Decode] ujson.loads(src)
yajl 0.3.5
  [Encode] yajl.dumps(d)
  [Decode] yajl.loads(src)
msgpack-0.6.2
  [Encode] msgpack.dumps(d)
  [Decode] msgpack.loads(src)

Encoding Test (15000 loops)
Package              Seconds
-----------------  ---------
pickle (binary-2)   0.141743
pickle (binary-3)   0.149022
pickle (binary-4)   0.151039
msgpack-0.6.2       0.155419
ujson-1.35          0.180287
yajl 0.3.5          0.388522
json                0.440383
simplejson-3.17.0   0.484677
pickle (ascii)      0.661336

Decoding Test (15000 loops)
Package              Seconds
-----------------  ---------
msgpack-0.6.2       0.136735
pickle (binary-4)   0.172758
pickle (ascii)      0.173402
pickle (binary-3)   0.175427
pickle (binary-2)   0.178137
ujson-1.35          0.206923
json                0.328846
yajl 0.3.5          0.438112
simplejson-3.17.0   0.475276

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment