jdk中的MappedByteBuffer非常类似linux中的mmap将文件映射到虚拟内存,这样可以利用操作系统自带的虚拟内存实现来
提高io效率, 很多文档中提到这一点,具体大家可以自行测试。
MappedByteBuffer有3种映射模式:
MapMode.READ_ONLY:只读映射
MapMode.READ_WRITE:读写映射
MapMode.PRIVATE:copy on write映射,一旦需要改变原有buffer,拷贝一份原来的
本文通过一种测试代码来理解MappedByteBuffer使用MapMode.READ_WRITE以及MapMode.PRIVATE,特别是MAP_MODE.PRIVATE模式
public class MMapTest {
private static FileChannel channel;
@BeforeClass
public static void runBeforeClass() throws FileNotFoundException {
String path = MMapTest.class.getResource("/test.txt").getPath();
File file = new File(path);
channel = new RandomAccessFile(file, "rw").getChannel();
}
@AfterClass
public static void runAfterClass() throws IOException {
channel.close();
}
@Before
public void runBeforeEveryTest() throws IOException {
ByteBuffer temp = ByteBuffer.allocate(100);
temp.put("12345".getBytes());
temp.flip();
channel.truncate(5);
channel.write(temp);
}
/**
*
* @author weip.pengw
* @time 2012-3-19 下午02:53:21
* @param buffer
* @throws Exception
*/
public static String dumpBuffer(ByteBuffer buffer) throws Exception {
StringBuffer sb = new StringBuffer();
int nulls = 0;
int limit = buffer.limit();
for (int i = 0; i < limit; i++) {
char c = (char) buffer.get(i);
if (c == '\u0000') {
nulls++;
continue;
}
if (nulls != 0) {
sb.append("|[" + nulls + " nulls]|");
nulls = 0;
}
sb.append(c);
}
return sb.toString();
}
/**
*
* @author weip.pengw
* @throws Exception
* @time 2012-3-19 下午02:45:28
*/
@Test
public void testCopyOnWrite() throws Exception {
// 产生一个文件,文件跨内存页
ByteBuffer temp = ByteBuffer.allocate(100);
temp.put("This is the file content".getBytes());
temp.flip();
channel.write(temp, 0);
temp.clear();
temp.put("This is more file content".getBytes());
temp.flip();
channel.write(temp, 8192);
// 产生两个MAPFILE
MappedByteBuffer rw = channel.map(FileChannel.MapMode.READ_WRITE, 0,
channel.size());
MappedByteBuffer cow = channel.map(FileChannel.MapMode.PRIVATE, 0,
channel.size());
// 测试之前
assertThat(
dumpBuffer(rw),
is("This is the file content|[8168 nulls]|This is more file content"));
assertThat(
dumpBuffer(cow),
is("This is the file content|[8168 nulls]|This is more file content"));
// 测试step1,修改rw前几个字节
rw.position(0);
rw.put("RW".getBytes());
rw.force();
assertThat(
dumpBuffer(rw),
is("RWis is the file content|[8168 nulls]|This is more file content"));
assertThat(
dumpBuffer(cow),
is("RWis is the file content|[8168 nulls]|This is more file content"));
// 测试step2,修改cow前几个字节,触发copy on write
cow.position(0);
cow.put("COW".getBytes());
assertThat(
dumpBuffer(rw),
is("RWis is the file content|[8168 nulls]|This is more file content"));
assertThat(
dumpBuffer(cow),
is("COWs is the file content|[8168 nulls]|This is more file content"));
// 测试step3,修改rw的最后几个字节,cow后面的字节反应了改变
rw.position(8192);
rw.put("RW".getBytes());
rw.force();
assertThat(
dumpBuffer(rw),
is("RWis is the file content|[8168 nulls]|RWis is more file content"));
assertThat(
dumpBuffer(cow),
is("COWs is the file content|[8168 nulls]|RWis is more file content"));
// 测试step4,修改cow的最后几个字节,再次触发copy on write
cow.position(8192);
cow.put("COW".getBytes());
assertThat(
dumpBuffer(rw),
is("RWis is the file content|[8168 nulls]|RWis is more file content"));
assertThat(
dumpBuffer(cow),
is("COWs is the file content|[8168 nulls]|COWs is more file content"));
// 测试step5,再次修改rw的前后几个字节,对cow没有了影响
rw.position(0);
rw.put("RW2".getBytes());
rw.position(8192);
rw.put("RW2".getBytes());
rw.force();
assertThat(
dumpBuffer(rw),
is("RW2s is the file content|[8168 nulls]|RW2s is more file content"));
assertThat(
dumpBuffer(cow),
is("COWs is the file content|[8168 nulls]|COWs is more file content"));
// cleanup
// channel.close();
}
重点看看代码中的测试方法testCopyOnWrite
1)产生一个较大空洞文件(本文操过8k),使得操作系统(代码在ubuntu10.04测试通过)无法在一页能映射该文件(跨内存页),这样才会有测试效果
2)将文件映射到两个MappedByteBuffer实例,一个是MapMode.READ_WRITE模式简称rw,一个是FileChannel.MapMode.PRIVATE模式简称cow
3) 测试之前,两个buff实例内容一致
4)第1步测试修改rw实例,由于没有触发copy on write,所以两个buff实例内容一致
5) 第2步测试修改cow实例的前几个字节,cow的修改只影响它本身,因为它修改的时候实际上重新拷贝了一份
6)第3步测试修改rw实例的后几个字节,居然在cow实例中能看到,之前不是说cow拷贝了一份吗?原因后面说
7)第4步测试修改cow实例的后几个字节,cow的修改只影响它本身
8)第5步测试修改rw实例的前后字节,发现都没反映在cow实例中,的确和6)有所不同。
这说明:
- 一开始没有修改cow,会共用rw的内存,
- 当cow修改了前几个字节,cow会拷贝内存,但明显cow实例是按内存页拷贝的,由于文件较长,所以只拷贝了前一个部分,前面的是自己的,后面是共享的
- 当cow修改了后几个字节,cow才拷贝后面的内存页,这样的话,bw对文件后面的修改,cow也就看不到了