Cglib 与 JDK动态代理的运行性能比较

都说 Cglib 创建的动态代理的运行性能比 JDK 动态代理能高出大概 10 倍，今日抱着怀疑精神验证了一下，发现情况有所不同，遂贴出实验结果，以供参考和讨论。

代码很简单，首先，定义一个 Test 接口，和一个实现 TestImpl 。Test 接口仅定义一个方法 test，对传入的 int 参数加 1 后返回。代码如下：

package my.test;

public interface Test {

    public int test(int i);

}

package my.test;

public class TestImpl implements Test{
    public int test(int i) {
        return i+1;
    }
}

然后，定义了三种代理的实现：装饰者模式实现的代理(decorator)，JDK 动态代理(dynamic proxy) 和 Cglib 动态代理 (cglib proxy)。代码如下：

package my.test;

public class DecoratorTest implements Test{
    private Test target;

    public DecoratorTest(Test target) {
        this.target = target;
    }

    public int test(int i) {
        return target.test(i);
    }
}

package my.test;

import java.lang.reflect.InvocationHandler;
import java.lang.reflect.Method;
import java.lang.reflect.Proxy;

public class DynamicProxyTest implements InvocationHandler {
    private Test target;

    private DynamicProxyTest(Test target) {
        this.target = target;
    }

    public static Test newProxyInstance(Test target) {
        return (Test) Proxy
                .newProxyInstance(DynamicProxyTest.class.getClassLoader(),
                        new Class<?>[] { Test.class },
                        new DynamicProxyTest(target));

    }

    public Object invoke(Object proxy, Method method, Object[] args)
            throws Throwable {
        return method.invoke(target, args);
    }
}

package my.test;

import java.lang.reflect.Method;

import net.sf.cglib.proxy.Enhancer;
import net.sf.cglib.proxy.MethodInterceptor;
import net.sf.cglib.proxy.MethodProxy;

public class CglibProxyTest implements MethodInterceptor {

    private CglibProxyTest() {
    }

    public static <T extends Test> Test newProxyInstance(Class<T> targetInstanceClazz){
        Enhancer enhancer = new Enhancer();
        enhancer.setSuperclass(targetInstanceClazz);
        enhancer.setCallback(new CglibProxyTest());
        return (Test) enhancer.create();
    }

    public Object intercept(Object obj, Method method, Object[] args,
            MethodProxy proxy) throws Throwable {
        return proxy.invokeSuper(obj, args);
    }

}

以 TestImpl 的调用耗时作为基准，对比通过其它三种代理进行调用的耗时。测试代码如下：

package my.test;

import java.util.LinkedHashMap;
import java.util.Map;

public class ProxyPerfTester {

    public static void main(String[] args) {
        //创建测试对象；
        Test nativeTest = new TestImpl();
        Test decorator = new DecoratorTest(nativeTest);
        Test dynamicProxy = DynamicProxyTest.newProxyInstance(nativeTest);
        Test cglibProxy = CglibProxyTest.newProxyInstance(TestImpl.class);

        //预热一下；
        int preRunCount = 10000;
        runWithoutMonitor(nativeTest, preRunCount);
        runWithoutMonitor(decorator, preRunCount);
        runWithoutMonitor(cglibProxy, preRunCount);
        runWithoutMonitor(dynamicProxy, preRunCount);

        //执行测试；
        Map<String, Test> tests = new LinkedHashMap<String, Test>();
        tests.put("Native   ", nativeTest);
        tests.put("Decorator", decorator);
        tests.put("Dynamic  ", dynamicProxy);
        tests.put("Cglib    ", cglibProxy);
        int repeatCount = 3;
        int runCount = 1000000;
        runTest(repeatCount, runCount, tests);
        runCount = 50000000;
        runTest(repeatCount, runCount, tests);
    }

    private static void runTest(int repeatCount, int runCount, Map<String, Test> tests){
        System.out.println(String.format("\n==================== run test : [repeatCount=%s] [runCount=%s] [java.version=%s] ====================", repeatCount, runCount, System.getProperty("java.version")));
        for (int i = 0; i < repeatCount; i++) {
            System.out.println(String.format("\n--------- test : [%s] ---------", (i+1)));
            for (String key : tests.keySet()) {
                runWithMonitor(tests.get(key), runCount, key);
            }
        }
    }

    private static void runWithoutMonitor(Test test, int runCount) {
        for (int i = 0; i < runCount; i++) {
            test.test(i);
        }
    }

    private static void runWithMonitor(Test test, int runCount, String tag) {
        long start = System.currentTimeMillis();
        for (int i = 0; i < runCount; i++) {
            test.test(i);
        }
        long end = System.currentTimeMillis();
        System.out.println("["+tag + "] Elapsed Time:" + (end-start) + "ms");
    }
}

测试用例分别在 jdk6、 jdk7、jdk8 下进行了测试，每次测试分别以 1,000,000 和 50,000,000 循环次数调用 test 方法，并重复3次。

• jdk6 下的测试结果如下：

==================== run test : [repeatCount=3] [runCount=1000000] [java.version=1.6.0_45] ====================

--------- test : [1] ---------
[Native   ] Elapsed Time:2ms
[Decorator] Elapsed Time:12ms
[Dynamic  ] Elapsed Time:31ms
[Cglib    ] Elapsed Time:31ms

--------- test : [2] ---------
[Native   ] Elapsed Time:7ms
[Decorator] Elapsed Time:7ms
[Dynamic  ] Elapsed Time:31ms
[Cglib    ] Elapsed Time:27ms

--------- test : [3] ---------
[Native   ] Elapsed Time:7ms
[Decorator] Elapsed Time:6ms
[Dynamic  ] Elapsed Time:23ms
[Cglib    ] Elapsed Time:29ms

==================== run test : [repeatCount=3] [runCount=50000000] [java.version=1.6.0_45] ====================

--------- test : [1] ---------
[Native   ] Elapsed Time:212ms
[Decorator] Elapsed Time:226ms
[Dynamic  ] Elapsed Time:1054ms
[Cglib    ] Elapsed Time:830ms

--------- test : [2] ---------
[Native   ] Elapsed Time:184ms
[Decorator] Elapsed Time:222ms
[Dynamic  ] Elapsed Time:1020ms
[Cglib    ] Elapsed Time:826ms

--------- test : [3] ---------
[Native   ] Elapsed Time:184ms
[Decorator] Elapsed Time:208ms
[Dynamic  ] Elapsed Time:979ms
[Cglib    ] Elapsed Time:832ms

　　测试结果表明：jdk6 下，在运行次数较少的情况下，jdk动态代理与 cglib 差距不明显，甚至更快一些；而当调用次数增加之后， cglib 表现稍微更快一些，然而仅仅是“稍微”好一些，远没达到 10 倍差距。

• jdk7 下的测试结果如下：

==================== run test : [repeatCount=3] [runCount=1000000] [java.version=1.7.0_60] ====================

--------- test : [1] ---------
[Native   ] Elapsed Time:2ms
[Decorator] Elapsed Time:12ms
[Dynamic  ] Elapsed Time:19ms
[Cglib    ] Elapsed Time:26ms

--------- test : [2] ---------
[Native   ] Elapsed Time:3ms
[Decorator] Elapsed Time:5ms
[Dynamic  ] Elapsed Time:17ms
[Cglib    ] Elapsed Time:20ms

--------- test : [3] ---------
[Native   ] Elapsed Time:4ms
[Decorator] Elapsed Time:4ms
[Dynamic  ] Elapsed Time:13ms
[Cglib    ] Elapsed Time:27ms

==================== run test : [repeatCount=3] [runCount=50000000] [java.version=1.7.0_60] ====================

--------- test : [1] ---------
[Native   ] Elapsed Time:208ms
[Decorator] Elapsed Time:210ms
[Dynamic  ] Elapsed Time:551ms
[Cglib    ] Elapsed Time:923ms

--------- test : [2] ---------
[Native   ] Elapsed Time:238ms
[Decorator] Elapsed Time:210ms
[Dynamic  ] Elapsed Time:483ms
[Cglib    ] Elapsed Time:872ms

--------- test : [3] ---------
[Native   ] Elapsed Time:217ms
[Decorator] Elapsed Time:208ms
[Dynamic  ] Elapsed Time:494ms
[Cglib    ] Elapsed Time:881ms

测试结果表明：jdk7 下，情况发生了逆转！在运行次数较少（1,000,000）的情况下，jdk动态代理比 cglib 快了差不多30%；而当调用次数增加之后(50,000,000)， 动态代理比 cglib 快了接近1倍。

接下来再看看jdk8下的表现如何。

• jdk8 下的测试结果如下：

==================== run test : [repeatCount=3] [runCount=1000000] [java.version=1.8.0_05] ====================

--------- test : [1] ---------
[Native   ] Elapsed Time:5ms
[Decorator] Elapsed Time:11ms
[Dynamic  ] Elapsed Time:27ms
[Cglib    ] Elapsed Time:52ms

--------- test : [2] ---------
[Native   ] Elapsed Time:4ms
[Decorator] Elapsed Time:6ms
[Dynamic  ] Elapsed Time:11ms
[Cglib    ] Elapsed Time:24ms

--------- test : [3] ---------
[Native   ] Elapsed Time:4ms
[Decorator] Elapsed Time:5ms
[Dynamic  ] Elapsed Time:9ms
[Cglib    ] Elapsed Time:26ms

==================== run test : [repeatCount=3] [runCount=50000000] [java.version=1.8.0_05] ====================

--------- test : [1] ---------
[Native   ] Elapsed Time:194ms
[Decorator] Elapsed Time:211ms
[Dynamic  ] Elapsed Time:538ms
[Cglib    ] Elapsed Time:965ms

--------- test : [2] ---------
[Native   ] Elapsed Time:194ms
[Decorator] Elapsed Time:214ms
[Dynamic  ] Elapsed Time:503ms
[Cglib    ] Elapsed Time:969ms

--------- test : [3] ---------
[Native   ] Elapsed Time:190ms
[Decorator] Elapsed Time:209ms
[Dynamic  ] Elapsed Time:495ms
[Cglib    ] Elapsed Time:939ms

测试结果表明：jdk8 下，延续了 JDK7 下的惊天大逆转！不过还观察另外有一个细微的变化，从绝对值来看 cglib 在 jdk8 下的表现似乎比 jdk7 还要差一点点，尽管只是一点点，但经过反复多次的执行仍然是这个趋势（注：这个趋势的结论并不严谨，只是捎带一提，如需得出结论还需进行更多样的对比实验）。

结论：从 jdk6 到 jdk7、jdk8 ，动态代理的性能得到了显著的提升，而 cglib 的表现并未跟上，甚至可能会略微下降。传言的 cglib 比 jdk动态代理高出 10 倍的情况也许是出现在更低版本的 jdk 上吧。

以上测试用例虽然简单，但揭示了 jdk 版本升级可能会带来一些新技术改变，会使我们以前的经验失效。放在真实业务场景下时，还需要按照实际情况进行测试后才能得出特定于场景的结论。

总之，实践出真知，还要与时俱进地去检视更新一些以往经验。

注：上述实验中 cglib 的版本是 3.1 。