String对我们来说太熟悉了,因为它无处不在,更因为用String可以描述这个世界几乎所有的东西,甚至于为了描述精确的数值都需要String出马(因为计算机眼中的二进制和人类眼中的十进制间总有那么点隔膜)。因为熟悉而变得简单,也容易被忽略。今天记录一下关于String的容易被忽略的两个问题。
- 字符串重用——节省内存
因为字符串太多,如果能够重用则能够节省很大的内存。首先看下面一个例子:String <span style="text-decoration: underline">string1</span> = "HELLOHELLO"; String <span style="text-decoration: underline">string2</span> = "HELLO" + "HELLO";
上面创建了几个字符串?1 or 2?后者是动态创建的,不过相信JVM可以对其直接优化的,因为编译时已经知道内容了,猜测是一个instance,即同一个char数组。Heapdump出来后观察果然是一个。
String string3 = args[0]+ args[1];
输入参数HELLO HELLO? 字符串变成几个?没错啊,是两个HELLOHELLO了。Dump heap后观察,果然是两个了。(其实不用dump healp,debug也可以看出来,string1和string3中的char[]指向地址是不一样的)。
依此延伸,可以而知由java反序列化而来的那些string也是不一样的。实例如下;
**public** **final** **static** **void** main(String[] args) **throws** Exception {
**new** StringDeserialized().testDescirialized();
}
**public** **void** testDescirialized() **throws** Exception {
String testString = "HELLOHELLO";
ObjectOutputStream dataOutStream = **new** ObjectOutputStream(**new** FileOutputStream("./stringdeserialized.data"));
**for** (**int** i = 0; i < 1000; i++)
dataOutStream.writeObject(testString);
dataOutStream.close();
List<String> readAgainList = **new** ArrayList<String>(100);
**for** (**int** i = 0; i < 100; i++) {
ObjectInputStream dataInputStream = **new** ObjectInputStream(**new** FileInputStream("./stringdeserialized.data"));
readAgainList.add((String) dataInputStream.readObject());
dataInputStream.close();
}
Thread._sleep_(Integer._MAX_VALUE_);
}
截图是heap dump出来的,有HELLOHELLO的个数有101个,占用的size>8080。对于JVM的内存使用可参考 http://www.javamex.com/tutorials/memory/object_memory_usage.shtml
问题来了,系统维护的数据大多是字符串信息,比如configserver,而很多的信息都是同一个字符串,那么反复的从网络序列化而来,占用多的Heap。当然自己可以写一个weak hashmap来维护,重用这些字符串。大家知道JVM中有String Pool,使用它无疑最好不过。查找String源码,发现intern()的注释如下:
* When the intern method is invoked, if the pool already contains a
* string equal to this <code>String</code> object as determined by
* the {@link #equals(Object)} method, then the string from the pool is
* returned. Otherwise, this <code>String</code> object is added to the
* pool and a reference to this <code>String</code> object is returned.
于是改变上面一行代码为:
readAgainList.add(((String) dataInputStream.readObject()).intern());
再次Heap dump分析如下,另外可以看出一个包含10个字符的String占用的Heap是80byte:
- 字符串序列化的速度
目前CS处理为了支持所谓的任意类型数据,CS采用了一个技巧,用Swizzle来保存java序列化后的byte类型,Server端无需反序列化就能保存任意类型的data;这样的坏处有两个:通用的Java序列化效率不高;协议不通用,对其他语言支持不行。因为目前的数据信息基本都是String类型,而对对String数据的专门处理,可以通过String内部的byte数组(UTF-8)类表示,这样也便于其他语言解析。可以考虑增加对publish(String)的支持。于是做了如下测试来比较对String不同serialize/deserialize的速率和大小。
结果是writeUTF最小最快,对于100char的String,差距是数量级的相当明显,虽然Swizzle使用了一个技巧,当对同一个swizzle instance多次传输时,无需重复的序列化。
PS:Swizzle简单的说就是把信息包装起来,然后把序列化的byte流缓存起来,这样如果同样的一个信息要推送/发送N次,就无能减少N-1次的序列化时间。
public class CompareSerialization {
**public** String generateTestData(**int** stringLength) {
Random random = **new** Random();
StringBuilder builder = **new** StringBuilder(stringLength);
**for** (**int** j = 0; j < stringLength; j++) {
builder.append((**char**) random.nextInt(127));
}
**return** builder.toString();
}
**public** **int** testJavaDefault(String data) **throws** Exception {
ObjectOutputStream outputStream = **null**;
ObjectInputStream inputStream = **null**;
**try** {
ByteArrayOutputStream byteArray = **new** ByteArrayOutputStream();
outputStream = **new** ObjectOutputStream(byteArray);
outputStream.writeObject(data);
outputStream.flush();
inputStream = **new** ObjectInputStream(**new** ByteArrayInputStream(byteArray.toByteArray()));
inputStream.readObject();
**return** byteArray.size();
}
**finally** {
outputStream.close();
inputStream.close();
}
}
**public** **int** testJavaDefaultBytes(String data) **throws** Exception {
ObjectOutputStream outputStream = **null**;
ObjectInputStream inputStream = **null**;
**try** {
ByteArrayOutputStream byteArray = **new** ByteArrayOutputStream();
outputStream = **new** ObjectOutputStream(byteArray);
outputStream.writeBytes(data);
outputStream.flush();
inputStream = **new** ObjectInputStream(**new** ByteArrayInputStream(byteArray.toByteArray()));
**byte**[] bytes = **new** **byte**[byteArray.size()];
inputStream.read(**new** **byte**[byteArray.size()]);
**new** String(bytes);
**return** byteArray.size();
}
**finally** {
outputStream.close();
inputStream.close();
}
}
**public** **int** testSwizzle(Swizzle data) **throws** Exception {
ObjectOutputStream outputStream = **null**;
ObjectInputStream inputStream = **null**;
**try** {
ByteArrayOutputStream byteArray = **new** ByteArrayOutputStream();
outputStream = **new** ObjectOutputStream(byteArray);
outputStream.writeObject(data);
outputStream.flush();
inputStream = **new** ObjectInputStream(**new** ByteArrayInputStream(byteArray.toByteArray()));
inputStream.readObject();
**return** byteArray.size();
}
**finally** {
outputStream.close();
inputStream.close();
}
}
**public** **int** testStringUTF(String data) **throws** Exception {
ObjectOutputStream outputStream = **null**;
ObjectInputStream inputStream = **null**;
**try** {
ByteArrayOutputStream byteArray = **new** ByteArrayOutputStream();
outputStream = **new** ObjectOutputStream(byteArray);
outputStream.writeUTF(data);
outputStream.flush();
inputStream = **new** ObjectInputStream(**new** ByteArrayInputStream(byteArray.toByteArray()));
inputStream.readUTF();
**return** byteArray.size();
}
**finally** {
outputStream.close();
inputStream.close();
}
}
**public** **final** **static** **void** main(String[] args) **throws** Exception {
CompareSerialization compare = **new** CompareSerialization();
String data = compare.generateTestData(Integer._parseInt_(args[0]));
Swizzle swizzle = **new** Swizzle(data);
System._out_.println("testJavaDefault size on networking:" + compare.testJavaDefault(data));
System._out_.println("testJavaDefaultBytes size on networking:" + compare.testJavaDefaultBytes(data));
System._out_.println("testStringUTF size on networking:" + compare.testStringUTF(data));
System._out_.println("testSwizzle size on networking:" + compare.testSwizzle(swizzle));
// warm up
**for** (**int** i = 0; i < 100; i++) {
compare.testJavaDefault(data);
compare.testJavaDefaultBytes(data);
compare.testStringUTF(data);
compare.testSwizzle(swizzle);
}
**long** startTime = System._currentTimeMillis_();
**for** (**int** i = 0; i < 10000; i++) {
compare.testJavaDefault(data);
}
**long** endTime = System._currentTimeMillis_();
System._out_.println("testJavaDefault using time:" + (endTime - startTime));
startTime = System._currentTimeMillis_();
**for** (**int** i = 0; i < 10000; i++) {
compare.testJavaDefaultBytes(data);
}
endTime = System._currentTimeMillis_();
System._out_.println("testJavaDefaultBytes using time:" + (endTime - startTime));
startTime = System._currentTimeMillis_();
**for** (**int** i = 0; i < 10000; i++) {
compare.testStringUTF(data);
}
endTime = System._currentTimeMillis_();
System._out_.println("testStringUTF using time:" + (endTime - startTime));
startTime = System._currentTimeMillis_();
**for** (**int** i = 0; i < 10000; i++) {
compare.testSwizzle(swizzle);
}
endTime = System._currentTimeMillis_();
System._out_.println("testSwizzle using time:" + (endTime - startTime));
}
本文来源于"阿里中间件团队播客",原文发表时间" 2011-11-18"