fastjson_benchmark

1. 测试环境

1.1 各个库的版本

fastjson2 2.0.15
fastjson1 1.0.83
jackson 2.13.4
gson 2.9.1
kryo 5.3.0
hessian 4.0.66

都是截止测试时间（2022-10-05）各个库的最新版本

1.2 服务器环境

测试的服务器基于阿里云最新代ECS，包括x64架构的Intel和AMD处理器，ARM架构的AltraMax和阿里云平头哥的倚天处理器。

cs.c7a.xlarge
ecs.c7a.xlarge
ecs.c6r.xlarge
ecs.g8m.xlarge

ecs.c7.xlarge

这个是阿里云当前代标准型ECS，处理器型号 Intel Xeon(Ice Lake) Platinum 8369B，4核，8G内存

ecs.c7a.xlarge

这个是阿里云当前代标准型ECS，处理器型号 AMD EPYC™ Milan 7T83，4核，8G内存

ecs.c6r.xlarge

这个是阿里云上售卖ARM处理器ECS，处理器型号 Ampere Altra / AltraMax，4核，8G内存

ecs.g8m.xlarge

这个是阿里云上售卖ARM处理器，处理器型号 Yitian 710，4核，16G内存。这个是阿里云平头哥的倚天710处理器，需要联系客服才能购买。

1.3 JDK版本

基于Oracle最新版本的Linux x64/aarch64的JDK版本，下载地址 https://www.oracle.com/java/technologies/

oracle-jdk1.8.0_341
oracle-jdk-11.0.16
oracle-jdk-17.0.4
oracle-jdk-18.0.2
oracle-jdk-19

1.4 测试代码以及运行方式

代码路径 https://github.com/alibaba/fastjson2/tree/2.0.15/benchmark/src/main/java/com/alibaba/fastjson2/benchmark/eishay
运行方式测试代码方式需要准备响应的JDK版本到Install目录下

git clone https://github.com/alibaba/fastjson2
cd fastjson2
git checkout 2.0.15
mvn clean install -Dmaven.test.skip
~/Install/jdk-1.8.0_341/bin/java -cp ~/git/fastjson2/benchmark/target/fastjson2-benchmarks.jar com.alibaba.fastjson2.benchmark.eishay.Eishay
~/Install/jdk-11.0.16/bin/java -cp ~/git/fastjson2/benchmark/target/fastjson2-benchmarks.jar com.alibaba.fastjson2.benchmark.eishay.Eishay
~/Install/jdk-17.0.4/bin/java -cp ~/git/fastjson2/benchmark/target/fastjson2-benchmarks.jar com.alibaba.fastjson2.benchmark.eishay.Eishay
~/Install/jdk-18.0.2/bin/java -cp ~/git/fastjson2/benchmark/target/fastjson2-benchmarks.jar com.alibaba.fastjson2.benchmark.eishay.Eishay
~/Install/jdk-19/bin/java -cp ~/git/fastjson2/benchmark/target/fastjson2-benchmarks.jar com.alibaba.fastjson2.benchmark.eishay.Eishay

2. 测试数据&结果分析

原始数据 https://github.com/alibaba/fastjson2/blob/2.0.15/docs/benchmark/benchmark_2.0.15_raw.md

2.1 EishayParseString

这个是最常用的场景，将JSON格式字符串反序列化为Java对象，这个场景在fastjson中的代码如下：

String str = "...";
Bean bean = JSON.parseJSONObject(str, Bean.class);

这个场景是fastjson1最强的场景；

在ecs.c7.xlarge-oracle-jdk1.8.0_341_x64环境中，只相当于只相当于fastjson2的76.07% ;jackson则只相当于fastjson2的39.15%; gson是fastjson2的33.18%，也就是说这个最常见的反序列化场景，fastjson2的性能差不多是jackson/gson的三倍。

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2 > fastjson1 > jackson > gson
jdk8  100%        76.07%      39.15%    33.18%
jdk11 100%        73.15%      39.20%    34.59%
jdk17 100%        91.18%      35.55%    33.98%
jdk18 100%        85.52%      36.21%    28.54%
jdk19 100%        86.91%      37.48%    29.09%

fastjson2在JDK11和JDK8的算法不同，导致这个场景下，fastjson2在JDK8的表现比JDK11下好，JDK17/18/19由于JDK自身性能优化，使得JDK17/18/19性能更好。JDK 18/19比JDK 17性能显著提升。

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2	fastjson1	jackson	gson
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	1320.023	1004.199 (76.07%)	516.82 (39.15%)	438.048 (33.18%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	1260.936	922.37 (73.15%)	494.333 (39.2%)	436.156 (34.59%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	1369.386	1248.583 (91.18%)	486.845 (35.55%)	465.337 (33.98%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	1395.646	1193.6 (85.52%)	505.428 (36.21%)	398.287 (28.54%)
ecs.c7.xlarge	oracle-jdk-19_x64	1388.353	1206.599 (86.91%)	520.411 (37.48%)	403.886 (29.09%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	1229.526	937.784 (76.27%)	477.09 (38.8%)	419.502 (34.12%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	1134.796	881.615 (77.69%)	470.933 (41.5%)	447.406 (39.43%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	1230.035	1263.187 (102.7%)	519.04 (42.2%)	516.882 (42.02%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	1244.473	1108.575 (89.08%)	457.552 (36.77%)	428.208 (34.41%)
ecs.c7a.xlarge	oracle-jdk-19_x64	1185.852	1177.811 (99.32%)	427.541 (36.05%)	400.906 (33.81%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	868.197	712.0 (82.01%)	376.827 (43.4%)	312.498 (35.99%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	849.48	736.375 (86.69%)	370.608 (43.63%)	316.399 (37.25%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	877.366	1002.943 (114.31%)	380.077 (43.32%)	315.595 (35.97%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	866.233	981.352 (113.29%)	364.075 (42.03%)	297.333 (34.32%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	879.06	980.02 (111.48%)	355.808 (40.48%)	293.269 (33.36%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	984.462	771.687 (78.39%)	375.51 (38.14%)	354.297 (35.99%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	1124.362	807.192 (71.79%)	413.95 (36.82%)	375.63 (33.41%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	1128.561	1114.467 (98.75%)	408.066 (36.16%)	353.178 (31.29%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	1136.232	1144.034 (100.69%)	391.845 (34.49%)	306.8 (27%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	1144.135	1133.377 (99.06%)	407.734 (35.64%)	350.49 (30.63%)

2.2 EishayParseStringPretty

这个场景是将格式化过的JSON格式字符串反序列化为Java对象，这个场景fastjson1的parse算法不擅长的。这个场景fastjson2表现仍然非常好。这个场景在ecs.c7.xlarge-oracle-jdk1.8.0_341_x64环境下，fastjson1的性能是fastjson2的27.15，jackson的性能是fastjson2的50.08%，gson是fastjson2的42.89%%.

这个场景在fastjson中的代码如下：

// 这里输入的是格式化过后的json字符串
String str = "{\n" +
        "\t\"id\":123\n" +
        "}";
Bean bean = JSON.parseJSONObject(str, Bean.class);

性能排序分别如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2 > jackson > gson  > fastjson1
jdk8  100%        50.08%    42.89%  28.58%
jdk11 100%        49.77%    45.50%  26.26%
jdk17 100%        48.29%    44.78%  31.33%
jdk18 100%        48.54%    42.21%  31.31%
jdk19 100%        46.65%    37.54%  28.57%

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2	fastjson1	jackson	gson
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	940.464	268.822 (28.58%)	480.396 (51.08%)	403.352 (42.89%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	907.039	238.149 (26.26%)	451.425 (49.77%)	412.661 (45.5%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	946.804	296.672 (31.33%)	457.174 (48.29%)	423.958 (44.78%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	933.883	292.434 (31.31%)	453.311 (48.54%)	394.177 (42.21%)
ecs.c7.xlarge	oracle-jdk-19_x64	988.807	282.481 (28.57%)	461.237 (46.65%)	371.236 (37.54%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	899.074	253.441 (28.19%)	437.011 (48.61%)	387.316 (43.08%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	788.548	243.077 (30.83%)	450.466 (57.13%)	431.675 (54.74%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	832.478	292.051 (35.08%)	465.183 (55.88%)	448.118 (53.83%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	797.438	271.997 (34.11%)	423.528 (53.11%)	405.21 (50.81%)
ecs.c7a.xlarge	oracle-jdk-19_x64	775.867	259.75 (33.48%)	412.325 (53.14%)	379.722 (48.94%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	685.863	208.277 (30.37%)	341.671 (49.82%)	280.223 (40.86%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	666.075	208.054 (31.24%)	348.135 (52.27%)	295.71 (44.4%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	662.112	240.315 (36.3%)	341.715 (51.61%)	289.012 (43.65%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	663.461	228.336 (34.42%)	334.155 (50.37%)	279.273 (42.09%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	665.034	221.462 (33.3%)	327.114 (49.19%)	271.69 (40.85%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	758.599	220.671 (29.09%)	349.148 (46.03%)	321.927 (42.44%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	739.847	208.959 (28.24%)	275.784 (37.28%)	348.357 (47.09%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	799.211	264.967 (33.15%)	372.393 (46.6%)	281.256 (35.19%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	673.976	255.994 (37.98%)	363.218 (53.89%)	291.881 (43.31%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	804.179	241.845 (30.07%)	360.28 (44.8%)	327.409 (40.71%)

2.3 EishayParseTreeString

这个场景是将JSON格式字符串反序列化为JSONObject，这也是一个最常见的场景之一，这个场景在fastjson中的代码如下：

String str = "...";
JSONObject jsonObject = JSON.parseJSONObject(str);

fastjson2在这个场景也表现了远超fastjson1/jackson/gson的性能。在不同的环境中，fastjson1和jackson都只能相当于fastjson2性能的40%~60%，gson则更差一些。

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2 > fastjson1 > jackson  > gson
jdk8  100%        49.11%      50.54%     33.08%
jdk11 100%        50.21%      53.41%     39.81%
jdk17 100%        50.87%      46.07%     28.98%
jdk18 100%        46.97%      43.04%     28.24%
jdk19 100%        49.03%      48.16%     30.10%

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2	fastjson1	jackson	gson
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	1024.392	503.042 (49.11%)	517.736 (50.54%)	338.902 (33.08%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	805.647	404.521 (50.21%)	430.305 (53.41%)	320.71 (39.81%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	1091.463	555.186 (50.87%)	502.859 (46.07%)	316.269 (28.98%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	1129.728	530.668 (46.97%)	486.195 (43.04%)	319.082 (28.24%)
ecs.c7.xlarge	oracle-jdk-19_x64	1049.116	514.414 (49.03%)	505.254 (48.16%)	315.743 (30.1%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	862.609	429.35 (49.77%)	504.816 (58.52%)	337.247 (39.1%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	795.942	399.425 (50.18%)	463.128 (58.19%)	346.302 (43.51%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	1046.206	524.425 (50.13%)	510.47 (48.79%)	385.776 (36.87%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	1012.796	527.705 (52.1%)	479.785 (47.37%)	370.695 (36.6%)
ecs.c7a.xlarge	oracle-jdk-19_x64	946.727	457.025 (48.27%)	487.359 (51.48%)	362.186 (38.26%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	585.244	300.751 (51.39%)	357.66 (61.11%)	260.348 (44.49%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	547.117	315.049 (57.58%)	348.928 (63.78%)	262.462 (47.97%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	614.709	357.832 (58.21%)	381.573 (62.07%)	259.427 (42.2%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	622.427	338.896 (54.45%)	360.341 (57.89%)	254.512 (40.89%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	603.085	337.843 (56.02%)	367.375 (60.92%)	253.413 (42.02%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	563.961	311.415 (55.22%)	334.18 (59.26%)	274.652 (48.7%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	664.295	363.316 (54.69%)	378.628 (57%)	300.685 (45.26%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	811.114	393.275 (48.49%)	428.312 (52.81%)	255.93 (31.55%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	825.188	378.532 (45.87%)	407.692 (49.41%)	286.672 (34.74%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	802.728	361.135 (44.99%)	432.181 (53.84%)	294.64 (36.7%)

2.4 EishayParseUTF8Bytes

这个是将UTF8格式的byte数组反序列化为Java对象，这个场景在缓存和RPC场景中常用。

这个场景在fastjson中的代码如下：

byte[] utf8Bytes = ...;
Bean bean = JSON.parseJSONObject(utf8Bytes, Bean.class);

这个场景fastjson2同样表现出了卓越的性能；在JDK8下，fastjson1和jackson的性能分别之后fastjson2的71.23%和53.76%; gson不直接支持输入utf8Bytes，需要先构造字符串，性能只有fastjson2的28.3%。

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2 > fastjson1 > jackson  > gson
jdk8  100%        70.10%      52.33%     27.40%
jdk11 100%        68.61%      51.24%     28.40%
jdk17 100%        80.81%      48.80%     26.27%
jdk18 100%        78.10%      47.57%     27.35%
jdk19 100%        77.79%      46.37%     26.03%

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2	fastjson1	jackson	gson
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	1122.673	787.029 (70.1%)	587.517 (52.33%)	307.605 (27.4%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	1095.4	751.501 (68.61%)	561.254 (51.24%)	311.108 (28.4%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	1186.373	958.667 (80.81%)	578.934 (48.8%)	311.608 (26.27%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	1164.931	909.776 (78.1%)	554.179 (47.57%)	318.601 (27.35%)
ecs.c7.xlarge	oracle-jdk-19_x64	1193.509	928.43 (77.79%)	553.385 (46.37%)	310.644 (26.03%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	865.765	735.685 (84.98%)	548.45 (63.35%)	313.526 (36.21%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	1035.644	696.77 (67.28%)	534.159 (51.58%)	356.966 (34.47%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	1044.923	873.254 (83.57%)	605.101 (57.91%)	358.599 (34.32%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	902.821	860.388 (95.3%)	569.01 (63.03%)	357.377 (39.58%)
ecs.c7a.xlarge	oracle-jdk-19_x64	948.099	813.541 (85.81%)	536.198 (56.56%)	345.395 (36.43%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	741.939	550.488 (74.2%)	457.982 (61.73%)	238.885 (32.2%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	764.826	583.751 (76.32%)	439.756 (57.5%)	252.718 (33.04%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	787.438	757.793 (96.24%)	425.687 (54.06%)	251.758 (31.97%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	781.739	737.323 (94.32%)	425.827 (54.47%)	247.366 (31.64%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	790.354	737.685 (93.34%)	419.229 (53.04%)	248.95 (31.5%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	840.027	658.393 (78.38%)	408.892 (48.68%)	244.654 (29.12%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	1026.276	684.56 (66.7%)	475.162 (46.3%)	283.144 (27.59%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	997.658	509.233 (51.04%)	429.713 (43.07%)	238.001 (23.86%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	1014.895	793.172 (78.15%)	459.629 (45.29%)	282.807 (27.87%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	1063.483	741.682 (69.74%)	453.99 (42.69%)	236.084 (22.2%)

2.5 EishayWriteString

这个是Java对象序列化成字符串，这个是序列化最常用的场景之一。这个场景在fastjson中的代码如下：

Bean bean = ...;
String str = JSON.toJSONString(bean);

在这个最常用的序列化场景，fastjson2表现出远超fastjson1/jackson/gson的性能；这个场景gson表现很差，只有fastjson2的1/5

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2 > jackson >   fastjson1  > gson
jdk8  100%        59.92%      37.64%       19.58%
jdk11 100%        60.15%      38.98%       18.50%
jdk17 100%        62.52%      37.94%       12.44%
jdk18 100%        53.64%      35.57%       12.10%
jdk19 100%        55.36%      36.74%       12.35%

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2	fastjson1	jackson	gson
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	1628.025	612.727 (37.64%)	975.517 (59.92%)	318.714 (19.58%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	1498.846	584.242 (38.98%)	901.584 (60.15%)	277.215 (18.5%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	1657.325	628.848 (37.94%)	1036.213 (62.52%)	206.112 (12.44%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	1681.715	598.113 (35.57%)	902.056 (53.64%)	203.571 (12.1%)
ecs.c7.xlarge	oracle-jdk-19_x64	1648.782	605.79 (36.74%)	912.733 (55.36%)	203.599 (12.35%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	1547.95	583.256 (37.68%)	905.673 (58.51%)	339.203 (21.91%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	1507.674	616.329 (40.88%)	900.552 (59.73%)	283.74 (18.82%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	1505.885	614.163 (40.78%)	916.059 (60.83%)	323.794 (21.5%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	1431.697	556.693 (38.88%)	813.363 (56.81%)	299.256 (20.9%)
ecs.c7a.xlarge	oracle-jdk-19_x64	1399.599	558.493 (39.9%)	803.304 (57.4%)	292.435 (20.89%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	1075.257	442.939 (41.19%)	569.243 (52.94%)	269.937 (25.1%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	1126.647	415.963 (36.92%)	612.196 (54.34%)	251.168 (22.29%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	1110.947	436.534 (39.29%)	634.159 (57.08%)	189.407 (17.05%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	1116.855	398.796 (35.71%)	575.919 (51.57%)	190.152 (17.03%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	1110.623	392.086 (35.3%)	597.186 (53.77%)	187.808 (16.91%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	1279.804	470.294 (36.75%)	637.485 (49.81%)	327.978 (25.63%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	1272.129	480.478 (37.77%)	680.737 (53.51%)	300.405 (23.61%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	1284.791	489.293 (38.08%)	620.568 (48.3%)	178.196 (13.87%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	1218.32	347.043 (28.49%)	554.791 (45.54%)	176.069 (14.45%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	1294.147	437.114 (33.78%)	528.188 (40.81%)	197.685 (15.28%)

2.6 EishayWriteUTF8Bytes

这个是Java对象序列化成UTF格式的byte数组，这个是最常用的场景之一，在缓存和RPC场景常用。gson不直接支持，需要先序列化为String再转换为UTF8格式的byte数组。

这个场景在fastjson中的代码如下：

Bean bean = ...;
byte[] utf8Bytes = JSON.toJSONBytes(bean);

在这个最常用的序列化场景，fastjson2表现出远超fastjson1/jackson/gson的性能；这个场景gson表现很差，只有fastjson2的1/5不到，在JDK 17下差距更大。

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2 > jackson >   fastjson1  > gson
jdk8  100%        58.10%      36.75%       19.04%
jdk11 100%        50.67%      32.61%       16.69%
jdk17 100%        49.10%      28.96%       10.49%
jdk18 100%        48.27%      27.80%       10.47%
jdk19 100%        49.35%      27.67%       10.34%

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2	fastjson1	jackson	gson
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	1542.518	566.808 (36.75%)	896.168 (58.1%)	293.629 (19.04%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	1659.306	541.071 (32.61%)	840.729 (50.67%)	277.005 (16.69%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	1964.89	569.104 (28.96%)	964.704 (49.1%)	206.146 (10.49%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	1940.193	539.343 (27.8%)	936.592 (48.27%)	203.104 (10.47%)
ecs.c7.xlarge	oracle-jdk-19_x64	1940.835	537.066 (27.67%)	957.826 (49.35%)	200.666 (10.34%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	1441.81	556.694 (38.61%)	796.929 (55.27%)	302.503 (20.98%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	1664.37	525.595 (31.58%)	805.813 (48.42%)	279.377 (16.79%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	1740.044	549.792 (31.6%)	846.86 (48.67%)	320.707 (18.43%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	1679.533	482.787 (28.75%)	772.376 (45.99%)	276.5 (16.46%)
ecs.c7a.xlarge	oracle-jdk-19_x64	1601.005	508.874 (31.78%)	792.242 (49.48%)	283.559 (17.71%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	1143.247	395.229 (34.57%)	564.72 (49.4%)	232.465 (20.33%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	1240.914	377.907 (30.45%)	557.564 (44.93%)	246.246 (19.84%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	1191.594	393.589 (33.03%)	590.895 (49.59%)	184.782 (15.51%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	1180.948	392.129 (33.2%)	583.529 (49.41%)	185.145 (15.68%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	1173.616	377.516 (32.17%)	554.243 (47.23%)	189.422 (16.14%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	1368.141	469.883 (34.34%)	660.369 (48.27%)	292.09 (21.35%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	1436.204	420.938 (29.31%)	432.712 (30.13%)	304.352 (21.19%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	1431.837	471.682 (32.94%)	676.724 (47.26%)	201.686 (14.09%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	1477.482	436.408 (29.54%)	519.773 (35.18%)	207.275 (14.03%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	1434.942	455.268 (31.73%)	617.044 (43%)	182.911 (12.75%)

2.7 EishayWriteStringTree

这个长江是JSONObject序列化成字符串，这个是序列化最常用的场景之一。这个场景在fastjson中的代码如下：

JSONObject jsonObject = ...;
String str = jsonObject.toJSONString();

在这个最常用的序列化场景，fastjson2表现出远超fastjson1/jackson/gson的性能；这个场景gson表现很差，只有fastjson2的1/3

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2 > jackson >   fastjson1  > gson
jdk8  100%        70.53%      62.26%       36.59%
jdk11 100%        76.55%      59.97%       29.99%
jdk17 100%        67.88%      64.60%       20.00%
jdk18 100%        71.11%      66.11%       20.59%
jdk19 100%        68.75%      64.34%       20.59%

aliyun ecs spec	jdk version	fastjson2	fastjson1	jackson	gson
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	1319.448	821.463 (62.26%)	930.632 (70.53%)	482.848 (36.59%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	1278.101	766.471 (59.97%)	978.442 (76.55%)	383.311 (29.99%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	1248.264	806.4 (64.6%)	847.335 (67.88%)	250.808 (20.09%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	1214.845	803.076 (66.11%)	863.923 (71.11%)	254.775 (20.97%)
ecs.c7.xlarge	oracle-jdk-19_x64	1232.079	792.473 (64.32%)	847.075 (68.75%)	253.668 (20.59%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	1168.834	796.983 (68.19%)	818.824 (70.05%)	426.338 (36.48%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	1103.669	760.369 (68.89%)	866.86 (78.54%)	354.722 (32.14%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	1094.031	819.069 (74.87%)	866.802 (79.23%)	428.089 (39.13%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	1050.412	789.896 (75.2%)	834.016 (79.4%)	409.174 (38.95%)
ecs.c7a.xlarge	oracle-jdk-19_x64	1083.27	719.327 (66.4%)	846.663 (78.16%)	406.81 (37.55%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	717.419	492.895 (68.7%)	574.373 (80.06%)	331.246 (46.17%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	768.666	505.407 (65.75%)	577.801 (75.17%)	281.126 (36.57%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	766.132	507.299 (66.22%)	545.404 (71.19%)	213.026 (27.81%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	765.887	514.326 (67.15%)	532.837 (69.57%)	215.826 (28.18%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	770.845	499.984 (64.86%)	564.606 (73.25%)	210.631 (27.32%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	967.262	584.524 (60.43%)	647.759 (66.97%)	413.261 (42.72%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	1044.727	626.387 (59.96%)	736.406 (70.49%)	360.298 (34.49%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	1010.276	628.604 (62.22%)	689.518 (68.25%)	213.635 (21.15%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	995.901	629.639 (63.22%)	685.909 (68.87%)	226.797 (22.77%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	991.672	619.929 (62.51%)	687.333 (69.31%)	226.537 (22.84%)

2.8 EishayWriteBinary

这个场景是将Java对象序列化为二进制的byte数组，是缓存序列化和RPC场景使用序列化协议的典型场景。

fastson2内置支持二进制格式jsonb，jsonb是fastjson2为了RPC场景设计的序列化和反序列化设置的高性能序列化协议。

jsonb有两种映射方式，将Java对象映射为类似JSONObject的KV格式，也可以映射类似JSONArray的数组格式。

fastjson2JSONBArrayMapping的代码如下：

Bean bean = ...;
byte[] jsonbBytes = JSONB.toBytes(bean, JSONWriter.Feature.BeanToArray);

fastjson2JSONB的代码如下：

Bean bean = ...;
byte[] jsonbBytes = JSONB.toBytes(bean);

fastjson2UTF8Bytes的代码如下：

Bean bean = ...;
byte[] utf8Bytes = JSON.toJSONBytes(bean);

这个场景下，jsonb的性能非常好，在JDK 11/17下性能更好，hessina和java内置序列化都表现很差。

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

      fastjson2JSONB > fastjson2UTF8Bytes> hessian > javaSerialize
jdk8  100%             71.28%              15.78%    10.34%
jdk11 100%             63.92%              13.19%     7.75%
jdk17 100%             53.13%               9.44%     6.63%
jdk18 100%             56.16%               9.50%     6.93%
jdk19 100%             56.09%               9.50%     6.23%

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2JSONB	fastjson2UTF8Bytes	hessian	javaSerialize
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	2148.347	1531.363 (71.28%)	339.071 (15.78%)	222.241 (10.34%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	2536.022	1620.997 (63.92%)	334.581 (13.19%)	196.662 (7.75%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	3375.077	1894.283 (56.13%)	318.638 (9.44%)	223.779 (6.63%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	3419.643	1920.304 (56.16%)	324.877 (9.5%)	236.868 (6.93%)
ecs.c7.xlarge	oracle-jdk-19_x64	3393.854	1903.59 (56.09%)	322.527 (9.5%)	211.602 (6.23%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	1862.576	1387.351 (74.49%)	382.54 (20.54%)	215.459 (11.57%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	2466.077	1708.984 (69.3%)	387.238 (15.7%)	224.947 (9.12%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	2779.975	1774.61 (63.84%)	385.193 (13.86%)	249.072 (8.96%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	2727.657	1624.962 (59.57%)	375.871 (13.78%)	238.437 (8.74%)
ecs.c7a.xlarge	oracle-jdk-19_x64	2735.581	1594.441 (58.29%)	353.076 (12.91%)	252.787 (9.24%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	1356.178	1128.731 (83.23%)	295.093 (21.76%)	180.776 (13.33%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	1951.14	1216.836 (62.37%)	312.456 (16.01%)	184.829 (9.47%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	1947.389	1210.047 (62.14%)	302.615 (15.54%)	181.923 (9.34%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	1906.848	1157.47 (60.7%)	300.276 (15.75%)	188.655 (9.89%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	1926.694	1194.592 (62%)	304.289 (15.79%)	180.769 (9.38%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	1654.976	1369.491 (82.75%)	346.682 (20.95%)	207.901 (12.56%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	2430.439	1458.663 (60.02%)	347.231 (14.29%)	219.179 (9.02%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	2306.656	1377.041 (59.7%)	347.787 (15.08%)	208.211 (9.03%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	2494.483	1444.006 (57.89%)	328.193 (13.16%)	220.988 (8.86%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	1617.92	1495.444 (92.43%)	340.406 (21.04%)	221.413 (13.69%)

2.9 EishayParseBinary

这个场景是将二进制的byte数组反序列化为Java对象，是缓存序列化和RPC场景使用序列化协议的典型场景。

fastjson2JSONBArrayMapping的代码如下：

byte[] jsonbBytes = JSONB.toBytes(bean, JSONWriter.Feature.BeanToArray);
Bean bean = JSONB.parseObject(jsonbBytes, Bean.class, JSONReader.Feature.SupportArrayToBean);

fastjson2JSONB的代码如下：

byte[] jsonbBytes = JSONB.toBytes(bean);
Bean bean = JSONB.parseObject(jsonbBytes, Bean.class);

fastjson2UTF8Bytes的代码如下：

byte[] utf8Bytes = JSON.toJSONBytes(bean);
Bean bean = JSON.parseObject(utf8Bytes, Bean.class);

性能排序如下：

// ecs.c7.xlarge-oracle-jdk1.8.0_341_x64

这个场景下，jsonb的性能非常好，在JDK 11/17下性能更好，hessina表现很差，java内置序列化都表现极差。

      fastjson2JSONB > fastjson2UTF8Bytes> hessian > javaSerialize
jdk8  100%             69.38%              17.00%    2.73%
jdk11 100%             48.46%              10.13%    1.94%
jdk17 100%             45.45%               9.43%    1.96%
jdk18 100%             43.51%               8.76%    1.83%
jdk19 100%             45.83%               8.47%    1.75%

下面是不同CPU不同JDK下的性能比较表格，第1列是阿里云机器规格，第2列是JDK的版本信息，第3列到第6列是不同库的JMH的Throughput数据，越大越好。第4列到第6列中的百分比是Throughput相比fastjson2的比，比如第三列的百分比是fastjson1/fastjson2的百分数表示。

aliyun ecs spec	jdk version	fastjson2JSONB	fastjson2UTF8Bytes	hessian	javaSerialize
ecs.c7.xlarge	oracle-jdk1.8.0_341_x64	1670.725	1159.104 (69.38%)	283.954 (17%)	45.549 (2.73%)
ecs.c7.xlarge	oracle-jdk-11.0.16_x64	2495.092	1209.058 (48.46%)	252.867 (10.13%)	48.526 (1.94%)
ecs.c7.xlarge	oracle-jdk-17.0.4_x64	2722.632	1237.478 (45.45%)	256.64 (9.43%)	53.233 (1.96%)
ecs.c7.xlarge	oracle-jdk-18.0.2_x64	2848.273	1239.177 (43.51%)	249.642 (8.76%)	52.094 (1.83%)
ecs.c7.xlarge	oracle-jdk-19_x64	2775.256	1271.933 (45.83%)	235.048 (8.47%)	48.616 (1.75%)
ecs.c7a.xlarge	oracle-jdk1.8.0_341_x64	1223.559	855.714 (69.94%)	228.647 (18.69%)	38.273 (3.13%)
ecs.c7a.xlarge	oracle-jdk-11.0.16_x64	2152.441	1077.133 (50.04%)	245.918 (11.43%)	44.802 (2.08%)
ecs.c7a.xlarge	oracle-jdk-17.0.4_x64	2767.375	1058.6 (38.25%)	246.159 (8.9%)	52.289 (1.89%)
ecs.c7a.xlarge	oracle-jdk-18.0.2_x64	2684.85	1033.888 (38.51%)	238.553 (8.89%)	48.603 (1.81%)
ecs.c7a.xlarge	oracle-jdk-19_x64	2550.169	965.844 (37.87%)	227.919 (8.94%)	38.476 (1.51%)
ecs.c6r.xlarge	oracle-jdk1.8.0_341_aarch64	1291.709	755.823 (58.51%)	188.712 (14.61%)	37.185 (2.88%)
ecs.c6r.xlarge	oracle-jdk-11.0.16_aarch64	1683.948	795.127 (47.22%)	177.161 (10.52%)	36.409 (2.16%)
ecs.c6r.xlarge	oracle-jdk-17.0.4_aarch64	1849.821	819.537 (44.3%)	205.808 (11.13%)	39.014 (2.11%)
ecs.c6r.xlarge	oracle-jdk-18.0.2_aarch64	1799.521	816.559 (45.38%)	172.351 (9.58%)	39.019 (2.17%)
ecs.c6r.xlarge	oracle-jdk-19_aarch64	1788.926	811.842 (45.38%)	172.142 (9.62%)	35.206 (1.97%)
ecs.g8m.xlarge	oracle-jdk1.8.0_341_aarch64	1478.455	857.263 (57.98%)	204.802 (13.85%)	42.098 (2.85%)
ecs.g8m.xlarge	oracle-jdk-11.0.16_aarch64	2075.398	1065.187 (51.32%)	194.874 (9.39%)	39.475 (1.9%)
ecs.g8m.xlarge	oracle-jdk-17.0.4_aarch64	2294.572	1086.885 (47.37%)	209.535 (9.13%)	38.386 (1.67%)
ecs.g8m.xlarge	oracle-jdk-18.0.2_aarch64	2220.858	1105.781 (49.79%)	189.986 (8.55%)	43.005 (1.94%)
ecs.g8m.xlarge	oracle-jdk-19_aarch64	2206.528	1150.675 (52.15%)	205.672 (9.32%)	39.964 (1.81%)

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fastjson_benchmark

1. 测试环境

1.1 各个库的版本

1.2 服务器环境

ecs.c7.xlarge

ecs.c7a.xlarge

ecs.c6r.xlarge

ecs.g8m.xlarge

1.3 JDK版本

1.4 测试代码以及运行方式

2. 测试数据&结果分析

2.1 EishayParseString

2.2 EishayParseStringPretty

2.3 EishayParseTreeString

2.4 EishayParseUTF8Bytes

2.5 EishayWriteString

2.6 EishayWriteUTF8Bytes

2.7 EishayWriteStringTree

2.8 EishayWriteBinary

2.9 EishayParseBinary

Clone this wiki locally