Comparison of python json libs in performance

I work a lot with json in python and never thought before about performance. I was using standard python json lib that goes right from the box, but today found out from one of my colleagues that there is way more faster lib called cjson. I got interested in this and googled a little bit. As a result i got a python script that shows which json lib is faster on Python. Here it is :

	import time
	import pickle
	import yajl
	try:
	import cjson
	except ImportError:
	cjson = None
	try:
	import simplejson
	except ImportError:
	simplejson = None
	try:
	import json
	except ImportError:
	json = None
	try:
	import ujson
	except ImportError:
	ujson = None
	default_data = {
	"name": "Foo",
	"type": "Bar",
	"count": 1,
	"info": {
	"x": 203,
	"y": 102,},}
	def ttt(f, data=None, x=100*1000):
	start = time.time()
	while x:
	x -= 1
	foo = f(data)
	return time.time()-start
	def profile(serial, deserial, data=None, x=100*1000):
	if not data:
	data = default_data
	squashed = serial(data)
	return (ttt(serial, data, x), ttt(deserial, squashed, x))
	def test(serial, deserial, data=None):
	if not data:
	data = default_data
	assert deserial(serial(data)) == data
	contenders = [
	('yajl', (yajl.Encoder().encode, yajl.Decoder().decode)),
	]
	if json:
	contenders.append(('json', (json.dumps, json.loads)))
	if cjson:
	contenders.append(('cjson', (cjson.encode, cjson.decode)))
	if simplejson:
	contenders.append(('simplejson', (simplejson.dumps, simplejson.loads)))
	if json:
	contenders.append(('stdlib json', (json.dumps, json.loads)))
	if ujson:
	contenders.append(('ultra json', (ujson.dumps, ujson.loads)))
	for name, args in contenders:
	test(*args)
	x, y = profile(*args)
	print("%-11s serialize: %0.3f deserialize: %0.3f total: %0.3f" % (
	name, x, y, x+y))

view raw compare_pytho_json_libs.py hosted with ❤ by GitHub

As a result i got this metrics :

yajl        serialize: 0.308  deserialize: 0.334  total: 0.642
json        serialize: 1.245  deserialize: 1.384  total: 2.629
cjson       serialize: 0.339  deserialize: 0.244  total: 0.583
stdlib json serialize: 1.019  deserialize: 1.396  total: 2.415
ultra json  serialize: 0.170  deserialize: 0.159  total: 0.328
[Finished in 7.0s]

And the winner is ultra json! :) If you want to clone this script for yourself - here's the link