1 概述

        在讨论动态代理机制时，一个不可避免的话题是性能。无论采用JDK动态代理还是CGLIB动态代理，本质上都是在原有目标对象上进行了封装和转换，这个过程需要消耗资源和性能。而JDK和CGLIB动态代理的内部实现过程本身也存在很大差异。下面将讨论两种动态代理机制对系统运行性能所带来的影响。

2 测试案例设计

        为了量化不同动态代理对性能的影响程度，将设计一个案例，在该案例中同样使用前面介绍的AccountService接口以及它的实现类AccountServiceImpl。将通过创建一定数量的AccountServiceImpl实例并调用它的doAccountTransaction()方法，触发动态代理机制。

        为了能够基于不同的代理机制来创建对象，需要引入Spring中一个非常有用的注解，即@Scope。这个注解可以用来设置Bean的作用域，还可以用来指定代理模式ScopedProxyMode，在Spring中，ScopedProxyMode是一个枚举，代码如下：

public enum ScopedProxyMode {DEFAULT,NO,INTERFACES,TARGET_CLASS;private ScopedProxyMode() {}
}

        这里的INTERFACES代表的就是JDK动态代理，而TARGET_CLASS使用的则是CGLIB动态代理。现在，创建两个配置类JDKProxyConfig和CGLIBProxyConfig，分别针对AccountServiceImpl指定两种不同的代理机制。JDKProxyConfig代码如下：

@Configuration
@EnableAspectJAutoProxy
public class JDKProxyConfig {@Bean@Scope(proxyMode = ScopedProxyMode.INTERFACES)public AccountService accountService(){return new AccountServiceImpl();}
}

        CGLIBProxyConfig代码如下：

@EnableAspectJAutoProxy
@Configuration
public class CGLIBProxyConfig {@Bean@Scope(proxyMode = ScopedProxyMode.TARGET_CLASS)public AccountService accountService(){return new AccountServiceImpl();}
}

        借助于这两个配置类，就可以通过AnnotationConfigApplicationContext这个基于注解配置的应用上下文对象来获取添加了不同代理机制的AccountServiceImpl对象。准备工作已经完成，编写一个测试用例来对不同代理机制的性能进行量化。代码如下：

public class ProxyTest {private static final Logger LOGGER = Logger.getLogger(ProxyTest.class);@Testpublic void testAopProxyPerformance() throws MinimumAccountException {int countObjects = 5000;AccountServiceImpl[] unproxiedClasses = new AccountServiceImpl[countObjects];for (int i = 0; i < countObjects; i++) {unproxiedClasses[i] = new AccountServiceImpl();}AccountService[] cglibProxyClasses = new AccountService[countObjects];AccountService accountService = null;for (int i = 0; i < countObjects; i++) {accountService = new AnnotationConfigApplicationContext(CGLIBProxyConfig.class).getBean(AccountService.class);cglibProxyClasses[i] = accountService;}AccountService[] jdkProxyClasses = new AccountService[countObjects];for (int i = 0; i < countObjects; i++) {accountService = new AnnotationConfigApplicationContext(JDKProxyConfig.class).getBean(AccountService.class);jdkProxyClasses[i] = accountService;}long noProxy = invokeTargetObjects(countObjects, unproxiedClasses);displayResults("NOProxy",noProxy);long jdkProxy = invokeTargetObjects(countObjects, jdkProxyClasses);displayResults("JDKProxy",jdkProxy);long cglibProxy = invokeTargetObjects(countObjects, cglibProxyClasses);displayResults("cglibProxy",cglibProxy);}private void displayResults(String label, long timeTook) {LOGGER.info(label + ": " + timeTook + "(ns) " + (timeTook / 1000000) + "(ms)");}private long invokeTargetObjects(int countObjects,AccountService[] classes) throws MinimumAccountException {long start = System.nanoTime();Account source = new Account(101,"Account1");Account dest = new Account(102,"Account2");for (int i = 0; i < countObjects; i++) {classes[i].doAccountTransaction(source, dest, 100);}long end = System.nanoTime();return end - start;}
}

        运行结果：

        可以看到，分别针对不使用代理、使用JDK代理和CGLIB代理的场景，创建了5000个AccountServiceImpl对象实例，并记录它们的创建时间。以上量化结果取决于不同的机器配置，但不影响得出结论。从结果中不难看出，JDK动态代理在性能上优于CGLIB动态代理，但相差并不大。 

  其他类的代码如下：

public class Account {private String accountName;private Integer accountNumber;public String getAccountName() {return accountName;}public void setAccountName(String accountName) {this.accountName = accountName;}public Integer getAccountNumber() {return accountNumber;}public void setAccountNumber(Integer accountNumber) {this.accountNumber = accountNumber;}public Account(Integer accountNumber, String accountName) {this.accountNumber = accountNumber;this.accountName = accountName;}
}

public interface AccountService {boolean doAccountTransaction(Account source, Account dest, int amount);
}

public class AccountServiceImpl implements AccountService {@Overridepublic boolean doAccountTransaction(Account source, Account dest, int amount) {return true;}}