JSON 性能测试之二 - wastjson 每秒 6GB 解析速度完胜 simdjson-java！

体验【开源中国】全新APP，一站式掌握开源技术动态，即刻下载，开启您的开源技术探索之旅！

前一篇JSON性能测试对比Fastjson2和DslJson，有评论提到和simdjson-java对比，那就测试走起.

说下simdjson-java这个库，之前也听说过，由于早前官网测试用的文本非常离谱，而且依赖JDK18+就没怎么关注，现在只要搜索simdjson（c++）基本上就能看到宣传最多的就是每秒千兆解析超越同类C++库甚至25倍的都有,不确定simdjson和simdjson-java性能差异有多大，下面只和simdjson-java做下对比测试。

测试环境: openJdk21 + window10 + i5  单线程

jvm参数:  -Xms3g -Xmx3g --add-opens=java.base/java.time=ALL-UNNAMED --add-modules=jdk.incubator.vector

maven 依赖截止当前最新版本

        <dependency>
            <groupId>org.simdjson</groupId>
            <artifactId>simdjson-java</artifactId>
            <version>0.3.0</version>
        </dependency>
        <dependency>
            <groupId>io.github.wycst</groupId>
            <artifactId>wast</artifactId>
            <version>0.0.19</version>
        </dependency>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-core</artifactId>
            <version>1.36</version>
        </dependency>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-generator-annprocess</artifactId>
            <version>1.36</version>
        </dependency>

场景测试一: 测试数据摘自simdjson-java官方(twitter.json内容已经相对正常多了)，数据大小:  631kb

import com.alibaba.fastjson2.JSON;
import com.alibaba.fastjson2.JSONObject;
import io.github.wycst.wast.common.utils.IOUtils;
import io.github.wycst.wast.json.JSONNode;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.Mode;
import org.openjdk.jmh.runner.Runner;
import org.openjdk.jmh.runner.RunnerException;
import org.openjdk.jmh.runner.options.Options;
import org.openjdk.jmh.runner.options.OptionsBuilder;
import org.openjdk.jmh.runner.options.TimeValue;
import org.simdjson.JsonValue;
import org.simdjson.SimdJsonParser;

import java.util.HashSet;
import java.util.Iterator;
import java.util.List;
import java.util.Set;

/**
 * @Date 2024/11/24 15:46
 * @Created by wangyc
 */
public class TwitterJmhTest {

    final static byte[] buffer;
    final static SimdJsonParser simdJsonParser = new SimdJsonParser();

    static {
        try {
            buffer = IOUtils.readBytes(TwitterJmhTest.class.getResourceAsStream("/data/json/twitter.json"));
            System.out.println("data size: " + (buffer.length >> 10) + "kb");
        } catch (Exception e) {
            throw new RuntimeException(e);
        }
    }

//    @Benchmark
//    public int parseAndSelect_Fastjson2() {
//        JSONObject jsonObject = (JSONObject) JSON.parse(buffer);
//        Set<String> defaultUsers = new HashSet<>();
//        Iterator<Object> tweets = jsonObject.getJSONArray("statuses").iterator();
//        while (tweets.hasNext()) {
//            JSONObject tweet = (JSONObject) tweets.next();
//            JSONObject user = (JSONObject) tweet.get("user");
//            if (user.getBoolean("default_profile")) {
//                defaultUsers.add(user.getString("screen_name"));
//            }
//        }
//        return defaultUsers.size();
//    }
//
//    @Benchmark
//    public int schemaBasedParseAndSelect_Fastjson2() {
//        Set<String> defaultUsers = new HashSet<>();
//        SimdJsonTwitter twitter = JSON.parseObject(buffer, SimdJsonTwitter.class);
//        for (SimdJsonStatus status : twitter.statuses) {
//            SimdJsonUser user = status.user;
//            if (user.default_profile) {
//                defaultUsers.add(user.screen_name);
//            }
//        }
//        return defaultUsers.size();
//    }

    @Benchmark
    public int parseAndSelect_WastJson() {
        List<JSONNode> users = JSONNode.collect(buffer, "/statuses/*/user");
        Set<String> defaultUsers = new HashSet<>();
        for (JSONNode user : users) {
            if (user.getChildValue("default_profile", boolean.class)) {
                defaultUsers.add(user.getChildValue("screen_name", String.class));
            }
        }
        return defaultUsers.size();
    }

    @Benchmark
    public int schemaBasedParseAndSelect_WastJson() {
        Set<String> defaultUsers = new HashSet<>();
        SimdJsonTwitter twitter = io.github.wycst.wast.json.JSON.parseObject(buffer, SimdJsonTwitter.class);
        for (SimdJsonStatus status : twitter.statuses) {
            SimdJsonUser user = status.user;
            if (user.default_profile) {
                defaultUsers.add(user.screen_name);
            }
        }
        return defaultUsers.size();
    }

    @Benchmark
    public int parseAndSelect_SimdjsonJava() {
        JsonValue simdJsonValue = simdJsonParser.parse(buffer, buffer.length);
        Set<String> defaultUsers = new HashSet<>();
        Iterator<JsonValue> tweets = simdJsonValue.get("statuses").arrayIterator();
        while (tweets.hasNext()) {
            JsonValue tweet = tweets.next();
            JsonValue user = tweet.get("user");
            if (user.get("default_profile").asBoolean()) {
                defaultUsers.add(user.get("screen_name").asString());
            }
        }
        return defaultUsers.size();
    }

    @Benchmark
    public int schemaBasedParseAndSelect_SimdjsonJava() {
        Set<String> defaultUsers = new HashSet<>();
        SimdJsonTwitter twitter = simdJsonParser.parse(buffer, buffer.length, SimdJsonTwitter.class);
        for (SimdJsonStatus status : twitter.statuses()) {
            SimdJsonUser user = status.user();
            if (user.default_profile()) {
                defaultUsers.add(user.screen_name());
            }
        }
        return defaultUsers.size();
    }

    record SimdJsonUser(boolean default_profile, String screen_name) {

    }

    record SimdJsonStatus(SimdJsonUser user) {

    }

    record SimdJsonTwitter(List<SimdJsonStatus> statuses) {

    }

    public static void main(String[] args) throws RunnerException {
        Options options = new OptionsBuilder()
                .include(TwitterJmhTest.class.getName())
                .mode(Mode.Throughput)
                .warmupIterations(1)
                .warmupTime(TimeValue.seconds(3))
                .measurementTime(TimeValue.seconds(3))
                .forks(1)
                .threads(1)
                .build();
        new Runner(options).run();
    }
}

测试结果（每秒运行次数）

Benchmark                                               Mode  Cnt     Score     Error  Units
TwitterJmhTest.parseAndSelect_SimdjsonJava             thrpt    5  1370.652 ±  37.171  ops/s
TwitterJmhTest.parseAndSelect_WastJson                 thrpt    5  1997.455 ± 275.386  ops/s
TwitterJmhTest.schemaBasedParseAndSelect_SimdjsonJava  thrpt    5  2092.163 ±  98.596  ops/s
TwitterJmhTest.schemaBasedParseAndSelect_WastJson      thrpt    5  2654.975 ± 102.240  ops/s

parseAndSelect：  WastJson 大约是 SimdjsonJava的1.45倍（解析速度wast 1.23g/s,  simdjson 0.84g/s）

schemaBasedParseAndSelect：WastJson大约是SimdjsonJava的1.26倍(解析速度wast 1.6g/s,  simdjson 1.28g/s)

以上是在simdjson-java官方提供数据下，WastJson都能完胜.

场景测试二:  全部为ascii编码的字节数组，大小1464kb 

测试代码

package com.jmh.test.simdjson;

import io.github.wycst.wast.json.JSON;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.Mode;
import org.openjdk.jmh.runner.Runner;
import org.openjdk.jmh.runner.RunnerException;
import org.openjdk.jmh.runner.options.Options;
import org.openjdk.jmh.runner.options.OptionsBuilder;
import org.openjdk.jmh.runner.options.TimeValue;
import org.simdjson.JsonValue;
import org.simdjson.SimdJsonParser;

import java.util.HashMap;
import java.util.Map;

/**
 * @Date 2024/11/27
 * @Created by wangyc
 */
public class AsciiSourceJmhTest {
    final static byte[] buffer;
    final static SimdJsonParser simdJsonParser = new SimdJsonParser();

    static {
        try {
            int length = 100000;
            StringBuilder value = new StringBuilder("a".repeat(length));
            Map map = new HashMap();
            map.put("a", value);
            map.put("b", value);
            map.put("c", value);
            map.put("d", value);
            map.put("e", value);
            map.put("f", value);
            map.put("g", value);
            map.put("h", value);
            map.put("i", value);
            map.put("j", value);
            map.put("k", value);
            map.put("l", value);
            map.put("m", value);
            map.put("n", value);
            map.put("o", value);
            buffer = io.github.wycst.wast.json.JSON.toJsonBytes(map);
            System.out.println("data size: " + (buffer.length >> 10) + "kb");
        } catch (Exception e) {
            throw new RuntimeException(e);
        }
    }

    @Benchmark
    public long ascii_WastJson() {
        long result = 0;
        Map map = (Map) JSON.parse(buffer);
        result += map.get("a").toString().length();
        result += map.get("b").toString().length();
        result += map.get("c").toString().length();
        result += map.get("d").toString().length();
        result += map.get("e").toString().length();
        result += map.get("f").toString().length();
        result += map.get("g").toString().length();
        result += map.get("h").toString().length();
        result += map.get("i").toString().length();
        result += map.get("j").toString().length();
        result += map.get("k").toString().length();
        result += map.get("l").toString().length();
        result += map.get("m").toString().length();
        result += map.get("n").toString().length();
        result += map.get("o").toString().length();
        return result;
    }

    @Benchmark
    public long ascii_SimdjsonJava() {
        long result = 0;
        JsonValue simdJsonValue = simdJsonParser.parse(buffer, buffer.length);
        result += simdJsonValue.get("a").asString().length();
        result += simdJsonValue.get("b").asString().length();
        result += simdJsonValue.get("c").asString().length();
        result += simdJsonValue.get("d").asString().length();
        result += simdJsonValue.get("e").asString().length();
        result += simdJsonValue.get("f").asString().length();
        result += simdJsonValue.get("g").asString().length();
        result += simdJsonValue.get("h").asString().length();
        result += simdJsonValue.get("i").asString().length();
        result += simdJsonValue.get("j").asString().length();
        result += simdJsonValue.get("k").asString().length();
        result += simdJsonValue.get("l").asString().length();
        result += simdJsonValue.get("m").asString().length();
        result += simdJsonValue.get("n").asString().length();
        result += simdJsonValue.get("o").asString().length();
        return result;
    }


    public static void main(String[] args) throws RunnerException {
        Options options = new OptionsBuilder()
                .include(AsciiSourceJmhTest.class.getName())
                .mode(Mode.Throughput)
                .warmupIterations(1)
                .warmupTime(TimeValue.seconds(3))
                .measurementTime(TimeValue.seconds(3))
                .forks(1)
                .threads(1)
                .build();
        new Runner(options).run();
    }


}

测试结果（每秒运行次数）

Benchmark                               Mode  Cnt     Score      Error  Units
AsciiSourceJmhTest.ascii_SimdjsonJava  thrpt    5  1231.578 ±  682.959  ops/s
AsciiSourceJmhTest.ascii_WastJson      thrpt    5  4212.839 ± 2349.629  ops/s

该场景下WastJson是Simdjson的3.4倍（解析速度 wast:  6.16GB/s, simdjson 1.8GB/s）

以上测试均为个人电脑环境下面的测试，感兴趣的也可以下载代码测试，如果有专业人士能测试不同平台的性能差异万分感谢！

测试代码仓库:   https://gitee.com/xiaoch0209/wast-jmh-test/tree/main-openjdk23/ (以上测试代码和测试结果都已上传)

询问AI