写入
package shareMemory;
import java.io.IOException;
import java.io.RandomAccessFile;
import java.nio.MappedByteBuffer;
import java.nio.channels.FileChannel;
import java.nio.channels.FileChannel.MapMode;
import java.nio.channels.FileLock;
import java.util.Iterator;
import java.util.Map.Entry;
import java.util.concurrent.Callable;
import java.util.concurrent.CompletionService;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ExecutorCompletionService;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.Future;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Writable;
import org.apache.hama.bsp.Combiner;
import org.apache.hama.graph.GraphJobMessage;
import org.apache.hama.util.ReflectionUtils;
import org.apache.hama.util.WritableUtils;
/**
* 模拟将同节点内部消息写在本地共享内存中以加速通信
*
* @author qiangliu
*
*/
public class ShareMemory {
public MappedByteBuffer mbb;
public static int bufferSize = 50 * 1024 * 1024; //内存
RandomAccessFile raf; // 共享内存对应文件
// public static String shareFile = "/opt/hama-0.7.1/shareMemory/shareMemory.txt";
public static String shareFile = "F:\\test\\shareMemory\\sharememory.txt";
ConcurrentHashMap<IntWritable, VertexMessage> vertexMessages = new ConcurrentHashMap<IntWritable, VertexMessage>(); //消息容器
private Combiner<Writable> combiner;
ShareMemory() {
try {
raf = new RandomAccessFile(shareFile, "rw");
final String combinerName = "org.apache.hama.examples.PageRank$PagerankCombiner";
// Class com = Class.forName(combinerName) ;
combiner = (Combiner<Writable>) ReflectionUtils
.newInstance(combinerName);
} catch (IOException e) {
e.printStackTrace();
} catch (ClassNotFoundException e) {
e.printStackTrace();
}
}
/**
* mock 10000 Pagerank消息,消息在写入共享内存时,发送消息时先保存在链表中,写入时用WritableUtils序列化后按字节写入
*/
public void generateMessages() {
//method 1 : 使用WritableUtils,后面建议改成在最后写入时再序列化,前面直接combine! 但是会存在当到同一个id的值很多且不能combine时就不好处理了!
for(int j = 1; j<3 ; j++) { //模拟3个super-step
for(int i = 0 ;i<1000; i++) {
DoubleWritable value = new DoubleWritable(i/j) ;
VertexMessage vMessage = new VertexMessage(new IntWritable(i), value) ;
if (combiner != null && vertexMessages.get(vMessage.getVertexId())!=null ) {
DoubleWritable combined = (DoubleWritable) combiner.combine(getIterableMessages(value)) ;
vMessage.setVertexVertexValue(combined) ; //更改合并后的值
vertexMessages.put(vMessage.getVertexId(), vMessage);
}
vertexMessages.put(vMessage.getVertexId(), vMessage) ;
}
}
}
public void close() {
try {
mbb.clear() ;
raf.close() ;
} catch (IOException e) {
e.printStackTrace();
}
}
public static void main(String[] args) {
ShareMemory shm = new ShareMemory() ;
long startTime = System.currentTimeMillis() ;
shm.generateMessages() ;
System.out.println("Generate Messages last :" + (System.currentTimeMillis() - startTime));
startTime = System.currentTimeMillis() ;
ExecutorService pool = Executors.newCachedThreadPool() ;
CompletionService<Boolean> exchangeResult = new ExecutorCompletionService<Boolean>(pool);
int destSize = 0 ;
destSize++ ;
exchangeResult.submit(shm.new MesssageShareSender()) ;
int count = 0 ;
while(count < destSize) {
Future<Boolean> f = exchangeResult.poll();
if(f == null) continue ;
count++;
}
System.out.println("Send Messages last :" + (System.currentTimeMillis() - startTime));
pool.shutdown() ;
shm.close() ;
}
public static Iterable<Writable> getIterableMessages(final Writable vertexValue) {
return new Iterable<Writable>() {
Writable value ;
@Override
public Iterator<Writable> iterator() {
value = vertexValue ;
return new Iterator<Writable>() {
int index = 1 ;
@Override
public boolean hasNext() {
return (index == 1) ? true : false;
}
@Override
public Writable next() {
index--;
return vertexValue;
}
@Override
public void remove() {
}
};
}
};
}
/**
* 同节点消息产生后,多线程发送给共享内存.两个问题: 1.为什么要用GraphJobMessage? 2. 直接写入快还是用流快?
* @author Administrator
*/
class MesssageShareSender implements Callable<Boolean>{
@Override
public Boolean call() throws Exception {
try {
FileChannel fc = raf.getChannel();
FileLock flock = fc.tryLock();
if(flock==null) {
Thread.sleep(10) ;
} else {
mbb = fc.map(MapMode.READ_WRITE, 0, ShareMemory.bufferSize); //因为写之前不知道需要映射多大共享内存,暂定50M
mbb.position(4) ; //预留一个int长度(4个字节)作为文件长度
int totalLength = 4 ; //最大2G
//模拟发送,最好像v0.6.4版本中,为每个任务的数据设置一个目录,所有发送到该任务的共享内存数据全部映射到这里!
Iterator<Entry<IntWritable, VertexMessage>> it = vertexMessages.entrySet()
.iterator();
while (it.hasNext()) {
Entry<IntWritable, VertexMessage> e = it.next();
it.remove();
byte[] serialized = WritableUtils.serialize(e.getValue().getVertexVertexValue()) ;
GraphJobMessage gjm = new GraphJobMessage(e.getValue().getVertexId(), serialized) ;
gjm.setVertexId(e.getKey());
gjm.setFlag(GraphJobMessage.VERTEX_FLAG);
byte[] message = WritableUtils.serialize(gjm) ;
int msgLen = message.length ;
mbb.putInt(msgLen) ; //这样快还是写入流读取,比如写到流里, DataOutput output output.toByteArray()
// System.out.println("Position : " + mbb.position());
mbb.put(message);
// System.out.println("Position : " + mbb.position());
totalLength = totalLength + msgLen +4 ;
}
mbb.putInt(0,mbb.position()) ; //补写长度
// System.out.println(" IsLoaded: " + mbb.isLoaded() + " Length: "+ totalLength +" Position: " + mbb.position());
}
} catch (Exception e) {
e.printStackTrace();
}
return true;
}
}
}
读取
package shareMemory;
import java.io.IOException;
import java.io.RandomAccessFile;
import java.nio.MappedByteBuffer;
import java.nio.channels.FileChannel;
import java.nio.channels.FileLock;
import java.nio.channels.FileChannel.MapMode;
import java.util.concurrent.ConcurrentHashMap;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.IntWritable;
import org.apache.hama.graph.GraphJobMessage;
import org.apache.hama.graph.GraphJobRunner;
import org.apache.hama.util.WritableUtils;
public class ReadingProcess {
byte[] buffer ; //缓存
public MappedByteBuffer mbb;
RandomAccessFile raf; // 共享内存对应文件
ConcurrentHashMap<IntWritable, VertexMessage> vertexMessages = new ConcurrentHashMap<IntWritable, VertexMessage>(); // 消息容器
public ReadingProcess() {
try {
raf = new RandomAccessFile(ShareMemory.shareFile, "rw");
FileChannel fc = raf.getChannel();
mbb = fc.map(MapMode.READ_ONLY, 0, fc.size()); // 映射的共享内存
mbb.load() ; // 预加载进内存
GraphJobRunner.VERTEX_ID_CLASS = IntWritable.class ;
GraphJobRunner.VERTEX_VALUE_CLASS = DoubleWritable.class ;
} catch (IOException e) {
e.printStackTrace();
}
}
public void close() {
try {
mbb.clear() ; //清空mbb
raf.close() ;
} catch (IOException e) {
e.printStackTrace();
}
}
public void readData() {
try {
FileChannel fc = raf.getChannel();
FileLock flock = fc.tryLock();
while(flock==null) { //轮询等待读取消息
Thread.sleep(10) ;
}
System.out.println("IsLoaded: " + mbb.isLoaded() +" position:"+ mbb.position());
int fileLength = mbb.getInt() ;
// buffer = new byte[fileLength] ;
// mbb.get(buffer) ; //本地消息缓存, 是否一次性读出?
while(mbb.position() < fileLength ) {
int msgLength = mbb.getInt() ;
if(msgLength>0) {
// System.out.println("Position : " + mbb.position());
byte[] message = new byte[msgLength] ;
mbb.get(message) ;
// System.out.println("Position : " + mbb.position());
GraphJobMessage gjm = new GraphJobMessage() ;
WritableUtils.deserialize(message, gjm) ;
if (!vertexMessages.containsKey(gjm.getVertexId())) {
DoubleWritable vertexValue = new DoubleWritable() ;
WritableUtils.deserialize(gjm.getValuesBytes(), vertexValue) ;
IntWritable vertexId = (IntWritable) gjm.getVertexId() ;
vertexMessages.put(vertexId, new VertexMessage(vertexId, vertexValue)) ;
} else {
System.out.println("Combine Error!");
}
}
}
// System.out.println("test ");
close() ;
} catch (Exception e) {
e.printStackTrace();
}
}
public static void main(String[] args) {
ReadingProcess rp = new ReadingProcess();
rp.readData() ;
}
}
分享到:
相关推荐
java的共享内存管理.基于MMF设计。封装了java.nio.MappedByteBuffer.在大流量实时业务系统时,可以极大的提高处理效率
深入浅出MappedByteBuffer
Bug ID 4724038 (fs) Add unmap method to MappedByteBuffer
mmfinvoker 这是一个简单的 java 库,它使用 nio.MappedByteBuffer 在内存映射文件上实现请求/响应功能。
本人初学c++,写了一个小软件,能把大文件分割问小文件,然后可以统国网络传输,到了网络另一端,再用此软件拼接! 希望用过的人能提宝贵意见! 13521825644 qq 362192302
返回说明 @return MappedByteBuffer 内存映射缓冲。 异常说明 throws 考虑异常 流程原理 调用实例 函数原型 private void readLog(MappedByteBuffer buffer, Vector<LogRecord> logins,Vector<LogRecord> ...
很多时候需要文件做数据交互,接收到文件后需要对文件解析成bean实体类,这里提供的是工具类,任意文件转任意实体都可以,只要简单的配置一下Class类,很实用
非易失性映射的字节缓冲将添加新的 JDK 特定文件映射模式,该模式允许 FileChannel API 用于创建引用非易失性内存(NVM)的 MappedByteBuffer 实例。 358:Helpful NullPointerExceptions 改进 ...
jdk14新特性:改进NullPointerExceptions,通过准确描述哪些变量为null...非易失性映射的字节缓冲将添加新的JDK特定文件映射模式,该模式允许FileChannel API用于创建引用非易失性内存(NVM)的MappedByteBuffer实例。
该库旨在将内存映射文件引入Java。 与已经存在的MappedByteBuffer相比,目标是更好地控制创建,修改和销毁。 它试图统一在不同操作系统上使用此类映射的接口,这意味着许多细节无法实现。 此外,在某些情况下,您...
当您只需要文档的一部分时(例如,您只需要信息字典或文档的页数),这将使内存占用量降至最低。 要读取的多个I / O实现。 SAMBox使用允许使用基于java.nio.channels.FileChannel , java.io.InputStream和java.nio...
百度地图开发java源码 tmp #inertiaSearch 2016年写的代码,现在觉得思路有很多提升的地方,但是毕竟努力过,还是贴在Readme.md 里面 ...对于原始的数据文件做内存映射,并做对应索引,所有索引做hash
a pratise of bigdata sorting,use some common util or class,like File,FileOutputStream,RandomAccessFile,HashMap,BufferedOutputStream,ByteBuffer,MappedByteBuffer,FileInputStream. as a newer of ...
the performance improvement is not so much as against using MappedByteBuffer. -> Otherwise I will totally use MappedByteBuffer 我无法在symbol.txt上进行验证。 单元测试:我找不到足够的时间来运行所有...
//MappedByteBuffer 将文件直接映射到内存 private int m_mbBufLen = 0; private int m_mbBufBegin = 0; private int m_mbBufEnd = 0; private String m_strCharsetName = "gbk";//文本格式 private Bitmap m_...
java redis源码杰里迪斯 Java的redis实现,与redis服务器相同。 特征 ...MappedByteBuffer 2.fixed unit size ,head 4 byte write in last item position, and then each item write in 4 byte with i
// 内存映射文件 private MappedByteBuffer mbb; // 单一模式实例 private static volatile IPSeeker instance = null; // 起始地区的开始和结束的绝对偏移 private long ipBegin, ipEnd; // 为提高...
MappedByteBuffer DirectByteBuffer HeapByteBuffer ShortBuffer IntBuffer LongBuffer FloatBuffer DoubleBuffer CharBuffer Selector选择器 Selector的作用就是配合一个线程来管理多个channel,获取这些channel上...
}读取ZIP文件注释,有两个版本的实现,这里使用的是 RandomAccessFile ,另一个版本使用的是 MappedByteBuffer ,经过测试,对于特别长的注释,使用内存映射文件读取性能要稍微好一些,对于特别短的注释(比如渠道名...