JSON serialization refactoring and optimization

[tomatosalat0]

Summary

• JSON.stringify now completes in about 60-50% of the time to complete the serialization. These numbers do apply when serializing plain objects without complex replacer or with custom toJSON methods. If you specify a spacer to beautify your generated JSONs, the serializer will take a little bit more time (see the table below), but still be faster compared to the previous revision.
• JSON.stringify now only allocates about 20% of the previously required memory during that process

Optimization

So long story short, the JSON serializer now uses a two StringBuilder for the complete operation instead of getting new ones quite often during the execution. One is used to build the JSON string itself and a second small one gets used when numbers get serialized.

The behavior of the serializer hasn't changed. The generated output will be the same as before.

Benchmarks

Setup

Like the in JSON.parse optimization process, I used several JSON files to measure the performance differences. Before executing the Benchmark, these JSON files are deserialized using the Jint JsonParser and then serialized back inside the benchmark method body. The benchmarks body basically looks like like this:

[Benchmark]
public void NewSerializerFormatted()
{
    JsonSerializer serializer = new JsonSerializer(_engine);
    serializer.Serialize(_toSerialize); // _toSerialize is the JsValue returned from JSON.parse
}

This time, I benched two calls: one without any spacer defined and one with spacer-value (serializer.Serialize(_toSerialize, JsValue.Undefined, new JsString(" "));)

These are the JSON-files used for the benchmarks

• https://github.com/sirthias/borer/tree/master/benchmarks/src/main/resources
• https://github.com/miloyip/nativejson-benchmark/tree/master/data
• https://github.com/algolia/examples/tree/master/demo-personalization (data.json)
• https://github.com/algolia/examples/tree/master/instant-search/instantsearch.js/dataset_import (bestbuy_dataset.json)

The JSON files for the test have a size between 2.7KB and 43.3MB (UTF-8 encoded).

Results

As like before, the table is huge. The relevant columns are "Ratio" and "Alloc Ratio".

• OldSerializerFormatted is the previous serializer, executed with two spaces for the space parameter
• OldSerializerUnformatted is the previous serializer, executed without any spacer
• NewSerializerXYZ is the same, but the new code

BenchmarkDotNet=v0.13.4, OS=Windows 10 (10.0.19042.2251/20H2/October2020Update)
AMD Ryzen 9 5950X, 1 CPU, 32 logical and 16 physical cores
.NET SDK=7.0.202
  [Host]     : .NET 6.0.15 (6.0.1523.11507), X64 RyuJIT AVX2
  DefaultJob : .NET 6.0.15 (6.0.1523.11507), X64 RyuJIT AVX2

Method	FileName	Mean	Error	StdDev	Ratio	RatioSD	Gen0	Gen1	Gen2	Allocated	Alloc Ratio
OldSerializerFormatted	algol(...)ation [31]	683,145.49 μs	7,773.648 μs	6,891.137 μs	1.00	0.00	49000.0000	13000.0000	-	988451.47 KB	1.00
NewSerializerFormatted	algol(...)ation [31]	388,183.85 μs	5,788.998 μs	5,415.033 μs	0.57	0.01	12000.0000	5000.0000	-	296073.65 KB	0.30
OldSerializerFormatted	australia-abc	348.42 μs	2.368 μs	2.215 μs	1.00	0.00	66.4063	15.6250	-	1090.33 KB	1.00
NewSerializerFormatted	australia-abc	198.81 μs	1.257 μs	1.176 μs	0.57	0.01	15.1367	3.6621	-	250.41 KB	0.23
OldSerializerFormatted	bestbuy_dataset	251,559.43 μs	3,728.770 μs	3,487.894 μs	1.00	0.00	20000.0000	6000.0000	500.0000	411267.09 KB	1.00
NewSerializerFormatted	bestbuy_dataset	167,054.41 μs	3,141.535 μs	3,361.407 μs	0.66	0.02	5333.3333	3333.3333	666.6667	121625.39 KB	0.30
OldSerializerFormatted	bitcoin	210.83 μs	1.919 μs	1.701 μs	1.00	0.00	45.1660	8.7891	-	739.25 KB	1.00
NewSerializerFormatted	bitcoin	115.01 μs	0.580 μs	0.542 μs	0.55	0.00	9.2773	1.5869	-	153.51 KB	0.21
OldSerializerFormatted	canada	97,960.85 μs	1,904.290 μs	3,233.629 μs	1.00	0.00	6833.3333	4833.3333	3000.0000	202056.34 KB	1.00
NewSerializerFormatted	canada	43,466.37 μs	749.058 μs	700.669 μs	0.44	0.02	1769.2308	1153.8462	615.3846	31033.46 KB	0.15
OldSerializerFormatted	citm_catalog	30,492.14 μs	106.101 μs	99.247 μs	1.00	0.00	3437.5000	2437.5000	1000.0000	51432.26 KB	1.00
NewSerializerFormatted	citm_catalog	11,833.41 μs	196.593 μs	183.893 μs	0.39	0.01	718.7500	484.3750	250.0000	10152.14 KB	0.20
OldSerializerFormatted	doj-blog	888.85 μs	11.427 μs	10.689 μs	1.00	0.00	454.1016	454.1016	454.1016	3175.7 KB	1.00
NewSerializerFormatted	doj-blog	486.27 μs	2.199 μs	1.950 μs	0.55	0.01	90.8203	90.8203	90.8203	634.08 KB	0.20
OldSerializerFormatted	eu-lobby-country	123.01 μs	1.251 μs	1.109 μs	1.00	0.00	23.1934	3.6621	-	381.02 KB	1.00
NewSerializerFormatted	eu-lobby-country	77.38 μs	0.522 μs	0.488 μs	0.63	0.01	5.8594	0.9766	-	97.77 KB	0.26
OldSerializerFormatted	eu-lobby-financial	783.66 μs	15.475 μs	15.198 μs	1.00	0.00	237.3047	237.3047	237.3047	1920.99 KB	1.00
NewSerializerFormatted	eu-lobby-financial	418.62 μs	3.060 μs	2.556 μs	0.54	0.01	47.3633	47.3633	47.3633	437.55 KB	0.23
OldSerializerFormatted	eu-lobby-repr	1,670.34 μs	22.008 μs	20.586 μs	1.00	0.00	554.6875	554.6875	554.6875	4215.33 KB	1.00
NewSerializerFormatted	eu-lobby-repr	826.59 μs	4.240 μs	3.759 μs	0.50	0.01	110.3516	110.3516	110.3516	918.64 KB	0.22
OldSerializerFormatted	github-events	1,284.82 μs	10.366 μs	9.697 μs	1.00	0.00	285.1563	142.5781	142.5781	4086.16 KB	1.00
NewSerializerFormatted	github-events	633.99 μs	3.525 μs	3.298 μs	0.49	0.00	71.2891	71.2891	71.2891	711.58 KB	0.17
OldSerializerFormatted	github-gists	725.52 μs	12.757 μs	11.309 μs	1.00	0.00	90.8203	90.8203	90.8203	1566.66 KB	1.00
NewSerializerFormatted	github-gists	348.45 μs	1.649 μs	1.462 μs	0.48	0.01	45.4102	45.4102	45.4102	412.41 KB	0.26
OldSerializerFormatted	inspe(...)yload [22]	470,566.69 μs	6,654.341 μs	5,898.900 μs	1.00	0.00	34000.0000	19000.0000	7000.0000	995097.57 KB	1.00
NewSerializerFormatted	inspe(...)yload [22]	114,521.19 μs	1,658.612 μs	1,470.316 μs	0.24	0.00	5200.0000	3000.0000	600.0000	110157.08 KB	0.11
OldSerializerFormatted	json-generator	116.39 μs	1.535 μs	1.436 μs	1.00	0.00	17.8223	1.5869	-	291.41 KB	1.00
NewSerializerFormatted	json-generator	72.50 μs	0.469 μs	0.439 μs	0.62	0.01	5.4932	0.8545	-	90.92 KB	0.31
OldSerializerFormatted	meteorites	10,006.58 μs	69.899 μs	61.964 μs	1.00	0.00	828.1250	546.8750	328.1250	9906.95 KB	1.00
NewSerializerFormatted	meteorites	3,317.38 μs	24.334 μs	21.572 μs	0.33	0.00	199.2188	199.2188	199.2188	2914.12 KB	0.29
OldSerializerFormatted	movies	93,649.51 μs	1,015.225 μs	949.642 μs	1.00	0.00	7500.0000	2833.3333	500.0000	136137.35 KB	1.00
NewSerializerFormatted	movies	52,278.88 μs	582.231 μs	544.619 μs	0.56	0.01	2000.0000	1000.0000	300.0000	38746.8 KB	0.28
OldSerializerFormatted	reddit-scala	1,511.92 μs	10.692 μs	9.478 μs	1.00	0.00	615.2344	615.2344	615.2344	4708.99 KB	1.00
NewSerializerFormatted	reddit-scala	644.47 μs	3.396 μs	3.011 μs	0.43	0.00	83.0078	83.0078	83.0078	727.96 KB	0.15
OldSerializerFormatted	rick-morty	238.72 μs	2.470 μs	2.190 μs	1.00	0.00	52.2461	10.4980	-	856.74 KB	1.00
NewSerializerFormatted	rick-morty	134.37 μs	0.377 μs	0.315 μs	0.56	0.01	11.9629	2.9297	-	196.64 KB	0.23
OldSerializerFormatted	temp-anomaly	92.85 μs	0.299 μs	0.265 μs	1.00	0.00	5.7373	0.2441	-	94.11 KB	1.00
NewSerializerFormatted	temp-anomaly	74.46 μs	0.403 μs	0.358 μs	0.80	0.00	1.7090	-	-	29.68 KB	0.32
OldSerializerFormatted	thai-cinemas	140.35 μs	1.399 μs	1.241 μs	1.00	0.00	25.3906	3.6621	-	418.29 KB	1.00
NewSerializerFormatted	thai-cinemas	76.83 μs	0.711 μs	0.665 μs	0.55	0.00	6.3477	0.8545	-	104.88 KB	0.25
OldSerializerFormatted	turkish	15,232.91 μs	297.883 μs	407.746 μs	1.00	0.00	1687.5000	1406.2500	984.3750	22091.77 KB	1.00
NewSerializerFormatted	turkish	6,821.00 μs	135.210 μs	296.789 μs	0.45	0.02	460.9375	367.1875	257.8125	4895.4 KB	0.22
OldSerializerFormatted	twitter	13,508.75 μs	252.575 μs	280.736 μs	1.00	0.00	1750.0000	1359.3750	968.7500	18560.21 KB	1.00
NewSerializerFormatted	twitter	3,776.76 μs	21.860 μs	19.379 μs	0.28	0.01	332.0313	332.0313	332.0313	3518.46 KB	0.19
OldSerializerFormatted	twitt(...)ponse [28]	119.14 μs	0.681 μs	0.637 μs	1.00	0.00	24.7803	3.2959	-	405.59 KB	1.00
NewSerializerFormatted	twitt(...)ponse [28]	69.48 μs	0.385 μs	0.341 μs	0.58	0.00	5.4932	0.8545	-	90.25 KB	0.22
OldSerializerFormatted	twitter_api_response	131.16 μs	0.817 μs	0.764 μs	1.00	0.00	27.0996	3.4180	-	444.95 KB	1.00
NewSerializerFormatted	twitter_api_response	76.96 μs	0.392 μs	0.367 μs	0.59	0.00	5.8594	0.8545	-	96.75 KB	0.22
OldSerializerUnformatted	algol(...)ation [31]	615,988.31 μs	8,281.685 μs	7,746.693 μs	1.00	0.00	39000.0000	10000.0000	-	787109.97 KB	1.00
NewSerializerUnformatted	algol(...)ation [31]	347,226.38 μs	3,207.867 μs	2,843.690 μs	0.56	0.01	10000.0000	4000.0000	-	242397.41 KB	0.31
OldSerializerUnformatted	australia-abc	314.00 μs	5.294 μs	4.952 μs	1.00	0.00	51.7578	10.2539	-	852 KB	1.00
NewSerializerUnformatted	australia-abc	183.62 μs	0.828 μs	0.734 μs	0.58	0.01	12.4512	2.9297	-	207.69 KB	0.24
OldSerializerUnformatted	bestbuy_dataset	236,860.49 μs	1,434.191 μs	1,119.722 μs	1.00	0.00	17666.6667	5333.3333	666.6667	354690.78 KB	1.00
NewSerializerUnformatted	bestbuy_dataset	137,904.51 μs	2,393.849 μs	2,239.208 μs	0.58	0.01	4750.0000	3250.0000	500.0000	107142.72 KB	0.30
OldSerializerUnformatted	bitcoin	195.37 μs	1.004 μs	0.939 μs	1.00	0.00	31.7383	4.6387	-	522.2 KB	1.00
NewSerializerUnformatted	bitcoin	109.46 μs	0.417 μs	0.348 μs	0.56	0.00	7.0801	0.9766	-	117.07 KB	0.22
OldSerializerUnformatted	canada	64,085.19 μs	1,082.357 μs	1,012.437 μs	1.00	0.00	3888.8889	2444.4444	1111.1111	100418.99 KB	1.00
NewSerializerUnformatted	canada	33,602.95 μs	658.807 μs	616.248 μs	0.52	0.01	937.5000	625.0000	250.0000	15737.25 KB	0.16
OldSerializerUnformatted	citm_catalog	22,492.49 μs	193.875 μs	161.894 μs	1.00	0.00	1937.5000	1031.2500	500.0000	28700.84 KB	1.00
NewSerializerUnformatted	citm_catalog	9,930.45 μs	175.368 μs	208.763 μs	0.44	0.01	484.3750	296.8750	125.0000	6955.75 KB	0.24
OldSerializerUnformatted	doj-blog	855.85 μs	4.920 μs	4.362 μs	1.00	0.00	454.1016	454.1016	454.1016	3098.47 KB	1.00
NewSerializerUnformatted	doj-blog	508.43 μs	5.068 μs	4.741 μs	0.59	0.01	90.8203	90.8203	90.8203	624.88 KB	0.20
OldSerializerUnformatted	eu-lobby-country	113.18 μs	0.981 μs	0.917 μs	1.00	0.00	19.5313	0.3662	-	320.68 KB	1.00
NewSerializerUnformatted	eu-lobby-country	66.37 μs	0.347 μs	0.324 μs	0.59	0.01	5.4932	0.7324	-	90.02 KB	0.28
OldSerializerUnformatted	eu-lobby-financial	695.48 μs	3.586 μs	3.179 μs	1.00	0.00	191.4063	191.4063	191.4063	1560.36 KB	1.00
NewSerializerUnformatted	eu-lobby-financial	391.93 μs	2.788 μs	2.608 μs	0.56	0.00	38.0859	38.0859	38.0859	372.06 KB	0.24
OldSerializerUnformatted	eu-lobby-repr	1,797.03 μs	33.367 μs	31.211 μs	1.00	0.00	498.0469	498.0469	498.0469	3766.65 KB	1.00
NewSerializerUnformatted	eu-lobby-repr	764.96 μs	4.191 μs	3.715 μs	0.43	0.01	99.6094	99.6094	99.6094	828.93 KB	0.22
OldSerializerUnformatted	github-events	1,070.18 μs	12.422 μs	11.620 μs	1.00	0.00	234.3750	117.1875	117.1875	3310.45 KB	1.00
NewSerializerUnformatted	github-events	565.97 μs	3.230 μs	2.864 μs	0.53	0.01	58.5938	58.5938	58.5938	586.54 KB	0.18
OldSerializerUnformatted	github-gists	634.22 μs	8.169 μs	7.641 μs	1.00	0.00	76.1719	76.1719	76.1719	1356.6 KB	1.00
NewSerializerUnformatted	github-gists	338.49 μs	1.660 μs	1.386 μs	0.53	0.01	38.0859	38.0859	38.0859	372.06 KB	0.27
OldSerializerUnformatted	inspe(...)yload [22]	277,221.96 μs	4,967.367 μs	11,708.669 μs	1.00	0.00	20000.0000	7000.0000	4000.0000	494897.48 KB	1.00
NewSerializerUnformatted	inspe(...)yload [22]	98,005.45 μs	1,188.018 μs	1,111.272 μs	0.34	0.02	3400.0000	1200.0000	200.0000	67977.36 KB	0.14
OldSerializerUnformatted	json-generator	113.40 μs	0.623 μs	0.583 μs	1.00	0.00	13.9160	1.0986	-	227.76 KB	1.00
NewSerializerUnformatted	json-generator	64.75 μs	0.302 μs	0.252 μs	0.57	0.00	4.8828	0.7324	-	81.45 KB	0.36
OldSerializerUnformatted	meteorites	5,128.21 μs	60.415 μs	56.512 μs	1.00	0.00	562.5000	406.2500	281.2500	7681.91 KB	1.00
NewSerializerUnformatted	meteorites	3,222.42 μs	21.024 μs	19.666 μs	0.63	0.01	140.6250	140.6250	140.6250	2365.84 KB	0.31
OldSerializerUnformatted	movies	88,612.62 μs	1,467.693 μs	1,301.072 μs	1.00	0.00	5500.0000	2500.0000	333.3333	99319.64 KB	1.00
NewSerializerUnformatted	movies	43,730.69 μs	488.550 μs	433.087 μs	0.49	0.01	1416.6667	666.6667	166.6667	28020.65 KB	0.28
OldSerializerUnformatted	reddit-scala	1,235.14 μs	18.078 μs	16.910 μs	1.00	0.00	468.7500	468.7500	468.7500	3646.09 KB	1.00
NewSerializerUnformatted	reddit-scala	629.95 μs	2.451 μs	2.293 μs	0.51	0.01	58.5938	58.5938	58.5938	593.59 KB	0.16
OldSerializerUnformatted	rick-morty	228.02 μs	1.661 μs	1.387 μs	1.00	0.00	41.5039	9.2773	-	679.91 KB	1.00
NewSerializerUnformatted	rick-morty	129.86 μs	0.560 μs	0.524 μs	0.57	0.00	9.7656	1.7090	-	161.45 KB	0.24
OldSerializerUnformatted	temp-anomaly	88.56 μs	0.332 μs	0.310 μs	1.00	0.00	4.7607	0.2441	-	79.75 KB	1.00
NewSerializerUnformatted	temp-anomaly	76.85 μs	0.333 μs	0.312 μs	0.87	0.00	1.3428	-	-	23.84 KB	0.30
OldSerializerUnformatted	thai-cinemas	126.86 μs	1.081 μs	1.011 μs	1.00	0.00	20.5078	2.6855	-	335.84 KB	1.00
NewSerializerUnformatted	thai-cinemas	70.74 μs	0.698 μs	0.619 μs	0.56	0.01	5.7373	0.8545	-	95.67 KB	0.28
OldSerializerUnformatted	turkish	15,284.55 μs	156.015 μs	138.303 μs	1.00	0.00	1593.7500	1375.0000	1000.0000	17471.31 KB	1.00
NewSerializerUnformatted	turkish	5,587.92 μs	110.290 μs	222.791 μs	0.36	0.02	375.0000	289.0625	203.1250	3968.95 KB	0.23
OldSerializerUnformatted	twitter	6,980.58 μs	77.399 μs	64.631 μs	1.00	0.00	1242.1875	1242.1875	1242.1875	13716.14 KB	1.00
NewSerializerUnformatted	twitter	3,270.05 μs	17.604 μs	16.467 μs	0.47	0.01	246.0938	246.0938	246.0938	2810.07 KB	0.20
OldSerializerUnformatted	twitt(...)ponse [28]	106.11 μs	0.972 μs	0.909 μs	1.00	0.00	18.1885	1.8311	-	299.16 KB	1.00
NewSerializerUnformatted	twitt(...)ponse [28]	65.65 μs	0.319 μs	0.283 μs	0.62	0.01	4.8828	0.7324	-	80.57 KB	0.27
OldSerializerUnformatted	twitter_api_response	121.02 μs	0.783 μs	0.732 μs	1.00	0.00	20.0195	2.1973	-	327.59 KB	1.00
NewSerializerUnformatted	twitter_api_response	71.13 μs	0.337 μs	0.299 μs	0.59	0.00	5.2490	0.3662	-	86.12 KB	0.26

[lahma]

Great work and performance improvement! 🚀