我有一个 JavaSE8 应用程序,用于并行处理大型数据集。我正在生成 1M 个对象,我想将其序列化为单个压缩文件。该文件将从网络应用程序下载/上传。 并行过程得到了很好的优化。然而,序列化/压缩是按顺序完成的,这是我的应用程序的瓶颈。
我测试了不同的解决方案:Kryo、ChronicleMap...我现在使用 Kryo 和 Bz2 压缩。它正在发挥作用。但性能还不够好。
我找不到任何进行并行序列化和压缩的解决方案。欢迎提供这方面的任何信息
最佳答案
实际上,如何并行或顺序处理数据集并不重要,因为在清晰的设计中 - 序列化始终是顺序操作(由于输出流、套接字等的顺序性质)操作并保留数据集加工。因此,如果您要序列化并将序列化的数据集放入文件、连接或原始内存中,您必须定义一个屏障,以保护数据免受并发竞争和意外修改的影响。
当然,在某些情况下,每个工作线程都会自行序列化数据,例如http服务器工作,但这里我们讨论的是并行处理并最终序列化的单个数据集。
所以,根据上面的说法,它应该是正确的答案代码。它使用标准的Java序列化+GZIP压缩。您可以轻松地替换此代码中的序列化和/或压缩,并针对您当前的解决方案进行基准测试。
package com.example.demo;
import java.io.*;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import java.util.zip.GZIPInputStream;
import java.util.zip.GZIPOutputStream;
import static java.lang.String.format;
public final class ParallelObjectsSerialization {
private static final int ONE_MILLION = 1_000_000;
private static final String SERIALIZE_FILE = "/tmp/out.bin";
public static void main(String[] args) throws IOException, ClassNotFoundException {
// List<Player> players = parallelGenerate1MPlayers();
List<Player> players = seqGenerate1MPlayers();
serialize(players);
players.clear();
players = deserialize();
}
private static List<Player> deserialize() throws IOException, ClassNotFoundException {
long started = System.currentTimeMillis();
List<Player> players = new ArrayList<>();
try (ObjectInputStream in = new ObjectInputStream(new GZIPInputStream(new FileInputStream(SERIALIZE_FILE)))) {
for (int i = 0; i < ONE_MILLION; i++) {
players.add((Player) in.readObject());
}
}
long time = System.currentTimeMillis() - started;
System.out.println(format("deserialization of %d objects took %d ms", players.size(), time));
return players;
}
private static final class Player implements Serializable {
private final String name;
private final int level;
private Player(String name, int level) {
this.name = name;
this.level = level;
}
}
private static List<Player> seqGenerate1MPlayers() {
long started = System.currentTimeMillis();
List<Player> players = new ArrayList<>(ONE_MILLION);
for (int i = 0; i < ONE_MILLION; i++) {
players.add(new Player(randomName(i), i));
}
long time = System.currentTimeMillis() - started;
System.out.println(format("sequential generating of %d objects took %d ms", players.size(), time));
return players;
}
private static List<Player> parallelGenerate1MPlayers() {
long started = System.currentTimeMillis();
Player[] players = new Player[ONE_MILLION];
Arrays.parallelSetAll(players, (i) -> new Player(randomName(i), i));
long time = System.currentTimeMillis() - started;
System.out.println(format("parallel generating of %d objects took %d ms", players.length, time));
return Arrays.asList(players);
}
private static void serialize(List<Player> players) throws IOException {
long started = System.currentTimeMillis();
try (ObjectOutputStream out = new ObjectOutputStream(new GZIPOutputStream(new FileOutputStream(SERIALIZE_FILE)))) {
for (Player player : players) {
out.writeObject(player);
}
}
long time = System.currentTimeMillis() - started;
System.out.println(format("serialization of %d objects took %d ms", players.size(), time));
}
private static String randomName(int seed) {
StringBuilder builder = new StringBuilder();
double chance = 30.0;
for (char c = 'a'; c <= 'z'; c++) {
if (Math.random() * 100.0 <= chance) {
builder.append(c);
if (builder.length() == 7) {
break;
}
}
}
if (builder.length() == 0) {
builder.append("unknown").append(seed);
}
return builder.toString();
}
}
关于Java并行序列化和压缩,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/47951886/