CLOVER🍀

That was when it all began.

Javaの圧縮・解凍ライブラリ（JDK標準GZIP、Commons Compress、snappy-java、LZ4 Java）を試す

Java

ちょっと圧縮・解凍ライブラリを調べていまして、簡単に以下あたりを比較したり、確認するための
下地を作ってみようかなと思いまして。

参考）
2016年のOSS圧縮ツール選択カタログ

Snappyとか名前をちらっと見かけたことがあるくらいだったので、こういう機会に1度見ておこうかと。

ピックしているものはけっこう適当だったり雰囲気だったりしますが、ご愛嬌。

なお、とりあえず確認にはテストコードを使うため、JUnitとAssertJをMaven依存関係に加えている
ものとします。

        <dependency>
            <groupId>junit</groupId>
            <artifactId>junit</artifactId>
            <version>4.12</version>
            <scope>test</scope>
        </dependency>
        <dependency>
            <groupId>org.assertj</groupId>
            <artifactId>assertj-core</artifactId>
            <version>3.6.2</version>
            <scope>test</scope>
        </dependency>

圧縮対象

Wikipedia日本語データのダンプの一部を使いましょう。
https://dumps.wikimedia.org/jawiki/20170320/

一部取ってきて

$ wget https://dumps.wikimedia.org/jawiki/20170320/jawiki-20170320-pages-articles1.xml-p000000001p000168815.bz2
$ bunzip2 jawiki-20170320-pages-articles1.xml-p000000001p000168815.bz2

適当に1K、1M、100M（全部およそ）くらいに分けます。

$ head -n 16 jawiki-20170320-pages-articles1.xml-p000000001p000168815 > 1K-file
$ head -n 7880 jawiki-20170320-pages-articles1.xml-p000000001p000168815 > 1M-file
$ head -n 1020000 jawiki-20170320-pages-articles1.xml-p000000001p000168815 > 100M-file

はい。

-rw-rw-r-- 1 xxxxx xxxxx  104222966  3月 30 00:31 100M-file
-rw-rw-r-- 1 xxxxx xxxxx       1069  3月 30 23:25 1K-file
-rw-rw-r-- 1 xxxxx xxxxx    1048529  3月 30 00:30 1M-file
-rw-rw-r-- 1 xxxxx xxxxx 1425464395  3月 23 00:31 jawiki-20170320-pages-articles1.xml-p000000001p000168815

こちらを使っていきましょう。ファイルは3つ作ったものの、エントリの量の都合上で最初は100Mのファイルを
使って書いていきます。

GZIP（JDK標準）

まずは、JDK標準で入っているGZIP。java.util.zipパッケージに入っているので、ライブラリ不要でそのまま使えます。

書いてみたサンプル。ファイルの中身をbyte配列にしてオンメモリで圧縮、それから展開して、最後にオリジナルのサイズと
圧縮後のサイズの比を取ります。
※以降のサンプルも、同じ内容で書いていきます
src/test/java/org/littlewings/compress/JdkGzipTest.java

package org.littlewings.compress;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.zip.GZIPInputStream;
import java.util.zip.GZIPOutputStream;

import org.junit.Ignore;
import org.junit.Test;

import static org.assertj.core.api.Assertions.assertThat;

public class JdkGzipTest {
    @Test
    public void compressDecompress100M() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        ByteArrayOutputStream compressBaos = new ByteArrayOutputStream();
        try (OutputStream gzip = new GZIPOutputStream(compressBaos)) {
            gzip.write(binary);
        }

        byte[] compressed = compressBaos.toByteArray();
        assertThat(compressed).hasSize(35534431);

        // 解凍
        ByteArrayOutputStream decompressBaos = new ByteArrayOutputStream();
        try (InputStream gzip = new GZIPInputStream(new ByteArrayInputStream(compressed))) {
            int b;
            while ((b = gzip.read()) != -1) {
                decompressBaos.write(b);
            }
        }

        byte[] decompressed = decompressBaos.toByteArray();
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.34, 0.35);
    }
}

圧縮後のデータサイズは約1/3になりました。

Commons Compress

続いて、Commons Compress。

Commons Compress – Overview

多くの圧縮形式をサポートしているようです。

ar, cpio, Unix dump, tar, zip, gzip, XZ, Pack200, bzip2, 7z, arj, lzma, snappy, DEFLATE, lz4 and Z files

http://commons.apache.org/proper/commons-compress/

サンプルも載っているので、それほど困らず使えると思います。
Commons Compress – Commons Compress User Guide

ただ、サンプルに載っているSnappyの書き出し、それからLZ4については、このエントリを書いている時点（1.13）では
まだ利用できません。

とりあえず、今回はGZIP、Bzip2、XZを試してみました。

利用するMaven依存関係は、こちら。

        <dependency>
            <groupId>org.apache.commons</groupId>
            <artifactId>commons-compress</artifactId>
            <version>1.13</version>
        </dependency>

GZIP。
src/test/java/org/littlewings/compress/CommonsGzipTest.java

package org.littlewings.compress;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.file.Files;
import java.nio.file.Paths;

import org.apache.commons.compress.compressors.gzip.GzipCompressorInputStream;
import org.apache.commons.compress.compressors.gzip.GzipCompressorOutputStream;
import org.junit.Ignore;
import org.junit.Test;

import static org.assertj.core.api.Assertions.assertThat;

public class CommonsGzipTest {
    @Test
    public void compressDecompress100M() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        ByteArrayOutputStream compressBaos = new ByteArrayOutputStream();
        try (OutputStream gzip = new GzipCompressorOutputStream(compressBaos)) {
            gzip.write(binary);
        }

        byte[] compressed = compressBaos.toByteArray();
        assertThat(compressed).hasSize(35534431);

        // 解凍
        ByteArrayOutputStream decompressBaos = new ByteArrayOutputStream();
        try (InputStream gzip = new GzipCompressorInputStream(new ByteArrayInputStream(compressed))) {
            int b;
            while ((b = gzip.read()) != -1) {
                decompressBaos.write(b);
            }
        }

        byte[] decompressed = decompressBaos.toByteArray();
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.34, 0.35);
    }
}

圧縮後のサイズは、JDK標準のGZIPと同じ（違ったら困りますが）。

続いて、Bzip2。
src/test/java/org/littlewings/compress/CommonsBzip2Test.java

package org.littlewings.compress;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.file.Files;
import java.nio.file.Paths;

import org.apache.commons.compress.compressors.bzip2.BZip2CompressorInputStream;
import org.apache.commons.compress.compressors.bzip2.BZip2CompressorOutputStream;
import org.apache.commons.compress.compressors.xz.XZCompressorInputStream;
import org.apache.commons.compress.compressors.xz.XZCompressorOutputStream;
import org.junit.Ignore;
import org.junit.Test;

import static org.assertj.core.api.Assertions.assertThat;

public class CommonsBzip2Test {
    @Test
    public void compressDecompress100M() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        ByteArrayOutputStream compressBaos = new ByteArrayOutputStream();
        try (OutputStream gzip = new BZip2CompressorOutputStream(compressBaos)) {
            gzip.write(binary);
        }

        byte[] compressed = compressBaos.toByteArray();
        assertThat(compressed).hasSize(25761240);

        // 解凍
        ByteArrayOutputStream decompressBaos = new ByteArrayOutputStream();
        try (InputStream gzip = new BZip2CompressorInputStream(new ByteArrayInputStream(compressed))) {
            int b;
            while ((b = gzip.read()) != -1) {
                decompressBaos.write(b);
            }
        }

        byte[] decompressed = decompressBaos.toByteArray();
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.24, 0.25);
    }
}

GZIPよりも高い圧縮率となります。1/4くらいのサイズになりました。

そして、最後にXZ。Commons CompressでXZを使うには、追加のライブラリが必要です。
XZ for Java

Maven依存関係としては、こちらを加えます。

        <dependency>
            <groupId>org.tukaani</groupId>
            <artifactId>xz</artifactId>
            <version>1.6</version>
        </dependency>

サンプル。
src/test/java/org/littlewings/compress/CommonsXzTest.java

package org.littlewings.compress;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.zip.GZIPOutputStream;

import org.apache.commons.compress.compressors.gzip.GzipCompressorInputStream;
import org.apache.commons.compress.compressors.xz.XZCompressorInputStream;
import org.apache.commons.compress.compressors.xz.XZCompressorOutputStream;
import org.junit.Ignore;
import org.junit.Test;

import static org.assertj.core.api.Assertions.assertThat;

public class CommonsXzTest {
    @Test
    public void compressDecompress100M() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        ByteArrayOutputStream compressBaos = new ByteArrayOutputStream();
        try (OutputStream gzip = new XZCompressorOutputStream(compressBaos)) {
            gzip.write(binary);
        }

        byte[] compressed = compressBaos.toByteArray();
        assertThat(compressed).hasSize(23685232);

        // 解凍
        ByteArrayOutputStream decompressBaos = new ByteArrayOutputStream();
        try (InputStream gzip = new XZCompressorInputStream(new ByteArrayInputStream(compressed))) {
            int b;
            while ((b = gzip.read()) != -1) {
                decompressBaos.write(b);
            }
        }

        byte[] decompressed = decompressBaos.toByteArray();
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.22, 0.23);
    }
}

GZIP、Bzip2よりもさらに小さくなります。

snappy-java

続いて、snappy-java。Snappyという圧縮形式のJava向けのライブラリです。

GitHub - xerial/snappy-java: Snappy compressor/decompressor for Java

JNIを使って、ネイティブコードを呼び出すような実装となっています。圧縮率はGZIPなどに比べると低いですが、
高速らしいです。

Maven依存関係は、こちらを追加。

        <dependency>
            <groupId>org.xerial.snappy</groupId>
            <artifactId>snappy-java</artifactId>
            <version>1.1.4-M3</version>
        </dependency>

使い方ですが、3種類のAPIがあり、簡易に使えるもの、Streamベースのもの、フレームを使ったものの3種類があります。
今回は、簡易的なものとStreamベースのものを使ってみたいと思います。

src/test/java/org/littlewings/compress/SnappySimplyTest.java

package org.littlewings.compress;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.file.Files;
import java.nio.file.Paths;

import org.apache.commons.compress.compressors.bzip2.BZip2CompressorInputStream;
import org.apache.commons.compress.compressors.bzip2.BZip2CompressorOutputStream;
import org.junit.Ignore;
import org.junit.Test;
import org.xerial.snappy.Snappy;

import static org.assertj.core.api.Assertions.assertThat;

public class SnappySimplyTest {
    @Test
    public void compressDecompress100M() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        byte[] compressed = Snappy.compress(binary);
        assertThat(compressed).hasSize(54457016);

        // 解凍
        byte[] decompressed = Snappy.uncompress(compressed);
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.52, 0.53);
    }
}

圧縮率は、確かに低いですね…半分にもなっていません。

続いて、Streamベースのもの。こちらは、大きなファイルの圧縮などに使うことを想定したAPIみたいです。
src/test/java/org/littlewings/compress/SnappyStreamTest.java

package org.littlewings.compress;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.file.Files;
import java.nio.file.Paths;

import org.junit.Ignore;
import org.junit.Test;
import org.xerial.snappy.SnappyInputStream;
import org.xerial.snappy.SnappyOutputStream;

import static org.assertj.core.api.Assertions.assertThat;

public class SnappyStreamTest {
    @Test
    public void compressDecompress100M() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        ByteArrayOutputStream compressBaos = new ByteArrayOutputStream();
        try (OutputStream gzip = new SnappyOutputStream(compressBaos)) {
            gzip.write(binary);
        }

        byte[] compressed = compressBaos.toByteArray();
        assertThat(compressed).hasSize(56361671);

        // 解凍
        ByteArrayOutputStream decompressBaos = new ByteArrayOutputStream();
        try (InputStream gzip = new SnappyInputStream(new ByteArrayInputStream(compressed))) {
            int b;
            while ((b = gzip.read()) != -1) {
                decompressBaos.write(b);
            }
        }

        byte[] decompressed = decompressBaos.toByteArray();
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.54, 0.55);
    }
}

ところで、これらのAPI注意点ですが、使うAPIによっては互換性がなかったりします。

例えば、簡易APIで圧縮したものをStreamベースのAPIで解凍できますが、その逆はできなかったりします。
src/test/java/org/littlewings/compress/SnappyApiCompatibilityTest.java

package org.littlewings.compress;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.nio.file.Files;
import java.nio.file.Paths;

import org.junit.Test;
import org.xerial.snappy.Snappy;
import org.xerial.snappy.SnappyInputStream;
import org.xerial.snappy.SnappyOutputStream;

import static org.assertj.core.api.Assertions.assertThat;
import static org.assertj.core.api.Assertions.assertThatThrownBy;

public class SnappyApiCompatibilityTest {
    @Test
    public void fromSimplyToStream() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        byte[] compressed = Snappy.compress(binary);
        assertThat(compressed).hasSize(54457016);

        // 解凍
        ByteArrayOutputStream decompressBaos = new ByteArrayOutputStream();
        try (InputStream gzip = new SnappyInputStream(new ByteArrayInputStream(compressed))) {
            int b;
            while ((b = gzip.read()) != -1) {
                decompressBaos.write(b);
            }
        }

        byte[] decompressed = decompressBaos.toByteArray();
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.52, 0.53);
    }

    @Test
    public void fromStreamToSimply() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        ByteArrayOutputStream compressBaos = new ByteArrayOutputStream();
        try (OutputStream gzip = new SnappyOutputStream(compressBaos)) {
            gzip.write(binary);
        }

        byte[] compressed = compressBaos.toByteArray();
        assertThat(compressed).hasSize(56361671);

        // 解凍
        assertThatThrownBy(() -> Snappy.uncompress(compressed))
                .isInstanceOf(IOException.class)
                .hasMessage("FAILED_TO_UNCOMPRESS(5)");
    }
}

フレームも含めて、APIの互換性は表にしてまとめられているので、こちらを参照しましょう。

snappy-java / Usage / Compatibility Notes

LZ4 Java

最後は、LZ4のJava向けのライブラリ。

GitHub - lz4/lz4-java: LZ4 compression for Java

Snappyよりも高速な圧縮形式らしいです。こちらのJava実装ですが、Snappyと同様JNIを使用してネイティブコードを
呼び出すようになっています。
（Pure Java実装も含まれていますが、最初にネイティブコードを使おうとするようになっています）

LZ4 Javaを使うのに必要なMaven依存関係は、こちら。

        <dependency>
            <groupId>net.jpountz.lz4</groupId>
            <artifactId>lz4</artifactId>
            <version>1.3.0</version>
        </dependency>

サンプルコード。
src/test/java/org/littlewings/compress/Lz4Test.java

package org.littlewings.compress;

import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Paths;

import net.jpountz.lz4.LZ4Compressor;
import net.jpountz.lz4.LZ4Factory;
import net.jpountz.lz4.LZ4FastDecompressor;
import org.junit.Test;

import static org.assertj.core.api.Assertions.assertThat;

public class Lz4Test {
    @Test
    public void compressDecompress100M() throws IOException {
        byte[] binary = Files.readAllBytes(Paths.get("../100M-file"));
        assertThat(binary).hasSize(104222966);

        // 圧縮
        LZ4Factory factory = LZ4Factory.fastestInstance();
        LZ4Compressor compressor = factory.fastCompressor();

        byte[] compressed = compressor.compress(binary);
        assertThat(compressed).hasSize(54393088);

        // 解凍
        LZ4FastDecompressor decompressor = factory.fastDecompressor();

        byte[] decompressed = decompressor.decompress(compressed, binary.length);
        assertThat(decompressed).hasSize(binary.length);

        assertThat((double) compressed.length / binary.length)
                .isBetween(0.52, 0.53);
    }
}

Snappyと同様、圧縮率自体はそれほど高くありません。

なお、ちょっと困ったこととして、解凍時に解凍後のサイズを渡す必要があるみたいなんですが…。

        byte[] decompressed = decompressor.decompress(compressed, binary.length);

GitHubのサンプルを見ると、確かに最初に圧縮前のサイズを保持していて、それを解凍時に渡すようにしています。

他のベンチマークとか見ていると、どうも圧縮後のファイルから計算で求められそうな雰囲気もあるのですが…果たして…？
https://github.com/ning/jvm-compressor-benchmark/blob/master/src/main/java/com/ning/jcbm/lz4/AbstractLz4Driver.java#L45-L55

とまあ、こんな感じで試してみました。

簡単にベンチマーク

と、ここまでAPIの簡単な使い方と圧縮後のサイズを見てきましたが、せっかくなので速度も簡単に見ておきます。

計測には、JMHを使いました。なんとなく。

OpenJDK: jmh

Maven依存関係には、こちらを追加。

        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-core</artifactId>
            <version>1.18</version>
        </dependency>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-generator-annprocess</artifactId>
            <version>1.18</version>
            <scope>provided</scope>
        </dependency>

また、Maven Shade Pluginも追加して、単一のJARファイルにしておきます。

    <build>
        <plugins>
            <plugin>
                <groupId>org.apache.maven.plugins</groupId>
                <artifactId>maven-shade-plugin</artifactId>
                <version>3.0.0</version>
                <executions>
                    <execution>
                        <phase>package</phase>
                        <goals>
                            <goal>shade</goal>
                        </goals>
                        <configuration>
                            <finalName>benchmark</finalName>
                            <transformers>
                                <transformer
                                        implementation="org.apache.maven.plugins.shade.resource.ManifestResourceTransformer">
                                    <mainClass>org.openjdk.jmh.Main</mainClass>
                                </transformer>
                            </transformers>
                            <filters>
                                <filter>
                                    <!--
                                        Shading signed JARs will fail without this.
                                        http://stackoverflow.com/questions/999489/invalid-signature-file-when-attempting-to-run-a-jar
                                    -->
                                    <artifact>*:*</artifact>
                                    <excludes>
                                        <exclude>META-INF/*.SF</exclude>
                                        <exclude>META-INF/*.DSA</exclude>
                                        <exclude>META-INF/*.RSA</exclude>
                                    </excludes>
                                </filter>
                            </filters>
                        </configuration>
                    </execution>
                </executions>
            </plugin>

        </plugins>
    </build>

ここまで使ってきた各種APIに対して、ファイルサイズ1K、1M、100Mに対して測定してみます。

参考）
他にもベンチマークの参考ということであれば、このあたりも見られた方がよいかもしれません。

GitHub - ning/jvm-compressor-benchmark: Benchmark suite for data compression library on the JVM

環境は、以下のとおり。

Java

$ java -version
java version "1.8.0_121"
Java(TM) SE Runtime Environment (build 1.8.0_121-b13)
Java HotSpot(TM) 64-Bit Server VM (build 25.121-b13, mixed mode)

Ubuntu Linux 14.04 LTS

$ uname -a
Linux xxxxx 3.13.0-79-generic #123-Ubuntu SMP Fri Feb 19 14:27:58 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

Core i7 CPU

$ cat /proc/cpuinfo | grep 'model name'
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz
model name	: Intel(R) Core(TM) i7-4710HQ CPU @ 2.50GHz

メモリは11G。

実行コマンドは、こんな感じ。
※絞ってますが、これでも2時間くらいかかるので…

$ java -jar benchmark.jar -wi 3 -i 3 -f 1 -bm avgt -tu ms

で、結果。
ソースコードはあとで載せます。結果がそこそこブレたりしたので、気になる方は手元で試されるとよいでしょう。

JDK標準GZIP

※結果は少し整形しています

# 圧縮
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.jdkGzipCompress_1K      avgt    3      0.022 ±    0.003  ms/op
CompressBenchmark.jdkGzipCompress_1M      avgt    3     58.445 ±   19.323  ms/op
CompressBenchmark.jdkGzipCompress_100M    avgt    3   6406.854 ± 1535.855  ms/op

#解凍
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.jdkGzipDecompress_1K    avgt    3      0.225 ±    0.097  ms/op
CompressBenchmark.jdkGzipDecompress_1M    avgt    3    210.369 ±   63.218  ms/op
CompressBenchmark.jdkGzipDecompress_100M  avgt    3  20953.244 ± 1796.023  ms/op
Commons Compress - GZIP
# 圧縮
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.commonsGzipCompress_1K       avgt    3      0.053 ±      0.717  ms/op
CompressBenchmark.commonsGzipCompress_1M       avgt    3     72.059 ±     29.865  ms/op
CompressBenchmark.commonsGzipCompress_100M     avgt    3   6600.498 ±  10184.787  ms/op

#解凍
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.commonsGzipDecompress_1K     avgt    3      0.290 ±      0.115  ms/op
CompressBenchmark.commonsGzipDecompress_1M     avgt    3    201.237 ±     44.347  ms/op
CompressBenchmark.commonsGzipDecompress_100M   avgt    3  20460.803 ±  11088.249  ms/op
Commons Compress - Bzip2
# 圧縮
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.commonsBzip2Compress_1K      avgt    3       1.739 ±       0.287  ms/op
CompressBenchmark.commonsBzip2Compress_1M      avgt    3     120.367 ±      11.274  ms/op
CompressBenchmark.commonsBzip2Compress_100M    avgt    3   11330.378 ±     227.484  ms/op

#解凍
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.commonsBzip2Decompress_1K    avgt    3       0.109 ±       0.063  ms/op
CompressBenchmark.commonsBzip2Decompress_1M    avgt    3      47.158 ±      15.537  ms/op
CompressBenchmark.commonsBzip2Decompress_100M  avgt    3    4815.504 ±     764.417  ms/op
Commons Compress - XZ
# 圧縮
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.commonsXzCompress_1K         avgt    3     13.180 ±    5.589  ms/op
CompressBenchmark.commonsXzCompress_1M         avgt    3    624.189 ±  266.903  ms/op
CompressBenchmark.commonsXzCompress_100M       avgt    3   95819.375 ±   15847.521  ms/op

#解凍
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.commonsXzDecompress_1K       avgt    3      1.245 ±    1.690  ms/op
CompressBenchmark.commonsXzDecompress_1M       avgt    3     65.512 ±   21.580  ms/op
CompressBenchmark.commonsXzDecompress_100M     avgt    3   5835.776 ± 1474.174  ms/op
Snappy - Simple
# 圧縮
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.snappySimplyCompress_1K      avgt    3      0.003 ±      0.001  ms/op
CompressBenchmark.snappySimplyCompress_1M      avgt    3      6.111 ±      0.382  ms/op
CompressBenchmark.snappySimplyCompress_100M    avgt    3   1583.644 ±  20934.928  ms/op

#解凍
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.snappySimplyDecompress_1K    avgt    3      0.001 ±      0.001  ms/op
CompressBenchmark.snappySimplyDecompress_1M    avgt    3      1.335 ±      0.237  ms/op
CompressBenchmark.snappySimplyDecompress_100M  avgt    3   1016.332 ±  27063.161  ms/op
Snappy - Stream
# 圧縮
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.snappyStreamCompress_1K      avgt    3       0.003 ±       0.002  ms/op
CompressBenchmark.snappyStreamCompress_1M      avgt    3       8.012 ±      53.825  ms/op
CompressBenchmark.snappyStreamCompress_100M    avgt    3     672.487 ±     414.928  ms/op

#解凍
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.snappyStreamDecompress_1K    avgt    3       0.005 ±       0.003  ms/op
CompressBenchmark.snappyStreamDecompress_1M    avgt    3       6.258 ±       3.308  ms/op
CompressBenchmark.snappyStreamDecompress_100M  avgt    3     811.926 ±    3204.878  ms/op
LZ4
# 圧縮
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.lz4Compress_1K               avgt    3      0.033 ±      0.488  ms/op
CompressBenchmark.lz4Compress_1M               avgt    3      4.618 ±     18.158  ms/op
CompressBenchmark.lz4Compress_100M             avgt    3  15281.482 ± 149076.550  ms/op

#解凍
Benchmark                                 Mode  Cnt      Score      Error  Units
CompressBenchmark.lz4Decompress_1K             avgt    3      0.024 ±      0.591  ms/op
CompressBenchmark.lz4Decompress_1M             avgt    3      0.990 ±      6.549  ms/op
CompressBenchmark.lz4Decompress_100M           avgt    3   3407.482 ±  59961.681  ms/op

まとめ

Snappy、LZ4が確かに高速です。GZIPとかと比べると、桁がひとつ以上違いますね…。ただ、圧縮率は劣るので、
そのあたりは用途と求める内容に応じてでしょうか。
※LZ4は100Mの時のばらつきがけっこう激しかったですが…

バランスを取るならGZIPだなぁという気もしてきました。

オマケ

ベンチマークを取った時のソースコードを載せておきます。
src/main/java/org/littlewings/compress/benchmark/CompressBenchmark.java

package org.littlewings.compress.benchmark;

import java.io.ByteArrayInputStream;
import java.io.ByteArrayOutputStream;
import java.io.IOException;
import java.io.InputStream;
import java.io.OutputStream;
import java.io.UncheckedIOException;
import java.nio.file.Files;
import java.nio.file.Paths;
import java.util.zip.GZIPInputStream;
import java.util.zip.GZIPOutputStream;

import net.jpountz.lz4.LZ4Compressor;
import net.jpountz.lz4.LZ4Factory;
import net.jpountz.lz4.LZ4FastDecompressor;
import org.apache.commons.compress.compressors.bzip2.BZip2CompressorInputStream;
import org.apache.commons.compress.compressors.bzip2.BZip2CompressorOutputStream;
import org.apache.commons.compress.compressors.gzip.GzipCompressorInputStream;
import org.apache.commons.compress.compressors.gzip.GzipCompressorOutputStream;
import org.apache.commons.compress.compressors.xz.XZCompressorInputStream;
import org.apache.commons.compress.compressors.xz.XZCompressorOutputStream;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.Scope;
import org.openjdk.jmh.annotations.Setup;
import org.openjdk.jmh.annotations.State;
import org.xerial.snappy.Snappy;
import org.xerial.snappy.SnappyInputStream;
import org.xerial.snappy.SnappyOutputStream;

@State(Scope.Benchmark)
public class CompressBenchmark {
    byte[][] binaries;

    byte[][] jdkGzipCompressed;

    byte[][] commonsGzipCompressed;

    byte[][] commonsBzip2Compressed;

    byte[][] commonsXzCompressed;

    byte[][] snappySimplyCompressed;

    byte[][] snappyStreamCompressed;

    byte[][] lz4Compressed;

    @Setup
    public void setup() throws IOException {
        binaries = new byte[][]{
                Files.readAllBytes(Paths.get("1K-file")),
                Files.readAllBytes(Paths.get("1M-file")),
                Files.readAllBytes(Paths.get("100M-file"))
        };

        jdkGzipCompressed = new byte[][]{
                jdkGzipCompress_1K(),
                jdkGzipCompress_1M(),
                jdkGzipCompress_100M()
        };

        commonsGzipCompressed = new byte[][]{
                commonsGzipCompress_1K(),
                commonsGzipCompress_1M(),
                commonsGzipCompress_100M()
        };

        commonsBzip2Compressed = new byte[][]{
                commonsBzip2Compress_1K(),
                commonsBzip2Compress_1M(),
                commonsBzip2Compress_100M()
        };

        commonsXzCompressed = new byte[][]{
                commonsXzCompress_1K(),
                commonsXzCompress_1M(),
                commonsXzCompress_100M()
        };

        snappySimplyCompressed = new byte[][]{
                snappySimplyCompress_1K(),
                snappySimplyCompress_1M(),
                snappySimplyCompress_100M()
        };

        snappyStreamCompressed = new byte[][]{
                snappyStreamCompress_1K(),
                snappyStreamCompress_1M(),
                snappyStreamCompress_100M()
        };

        lz4Compressed = new byte[][]{
                lz4Compress_1K(),
                lz4Compress_1M(),
                lz4Compress_100M()
        };
    }

    ///// JDK Gzip
    @Benchmark
    public byte[] jdkGzipCompress_1K() {
        return streamBasedCompress(binaries[0], is -> new GZIPOutputStream(is));
    }

    @Benchmark
    public byte[] jdkGzipCompress_1M() {
        return streamBasedCompress(binaries[1], is -> new GZIPOutputStream(is));
    }

    @Benchmark
    public byte[] jdkGzipCompress_100M() {
        return streamBasedCompress(binaries[2], is -> new GZIPOutputStream(is));
    }

    @Benchmark
    public byte[] jdkGzipDecompress_1K() {
        return streamBasedDecompress(jdkGzipCompressed[0], is -> new GZIPInputStream(is));
    }

    @Benchmark
    public byte[] jdkGzipDecompress_1M() {
        return streamBasedDecompress(jdkGzipCompressed[1], is -> new GZIPInputStream(is));
    }

    @Benchmark
    public byte[] jdkGzipDecompress_100M() {
        return streamBasedDecompress(jdkGzipCompressed[2], is -> new GZIPInputStream(is));
    }

    ///// Commons Compress Gzip
    @Benchmark
    public byte[] commonsGzipCompress_1K() {
        return streamBasedCompress(binaries[0], is -> new GzipCompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsGzipCompress_1M() {
        return streamBasedCompress(binaries[1], is -> new GzipCompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsGzipCompress_100M() {
        return streamBasedCompress(binaries[2], is -> new GzipCompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsGzipDecompress_1K() {
        return streamBasedDecompress(commonsGzipCompressed[0], is -> new GzipCompressorInputStream(is));
    }

    @Benchmark
    public byte[] commonsGzipDecompress_1M() {
        return streamBasedDecompress(commonsGzipCompressed[1], is -> new GzipCompressorInputStream(is));
    }

    @Benchmark
    public byte[] commonsGzipDecompress_100M() {
        return streamBasedDecompress(commonsGzipCompressed[2], is -> new GzipCompressorInputStream(is));
    }

    ///// Commons Compress Bzip2
    @Benchmark
    public byte[] commonsBzip2Compress_1K() {
        return streamBasedCompress(binaries[0], is -> new BZip2CompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsBzip2Compress_1M() {
        return streamBasedCompress(binaries[1], is -> new BZip2CompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsBzip2Compress_100M() {
        return streamBasedCompress(binaries[2], is -> new BZip2CompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsBzip2Decompress_1K() {
        return streamBasedDecompress(commonsBzip2Compressed[0], is -> new BZip2CompressorInputStream(is));
    }

    @Benchmark
    public byte[] commonsBzip2Decompress_1M() {
        return streamBasedDecompress(commonsBzip2Compressed[1], is -> new BZip2CompressorInputStream(is));
    }

    @Benchmark
    public byte[] commonsBzip2Decompress_100M() {
        return streamBasedDecompress(commonsBzip2Compressed[2], is -> new BZip2CompressorInputStream(is));
    }

    ///// Commons Compress Xz
    @Benchmark
    public byte[] commonsXzCompress_1K() {
        return streamBasedCompress(binaries[0], is -> new XZCompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsXzCompress_1M() {
        return streamBasedCompress(binaries[1], is -> new XZCompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsXzCompress_100M() {
        return streamBasedCompress(binaries[2], is -> new XZCompressorOutputStream(is));
    }

    @Benchmark
    public byte[] commonsXzDecompress_1K() {
        return streamBasedDecompress(commonsXzCompressed[0], is -> new XZCompressorInputStream(is));
    }

    @Benchmark
    public byte[] commonsXzDecompress_1M() {
        return streamBasedDecompress(commonsXzCompressed[1], is -> new XZCompressorInputStream(is));
    }

    @Benchmark
    public byte[] commonsXzDecompress_100M() {
        return streamBasedDecompress(commonsXzCompressed[2], is -> new XZCompressorInputStream(is));
    }

    ///// Snappy-Java Simply
    @Benchmark
    public byte[] snappySimplyCompress_1K() {
        return snappySimplyCompress(binaries[0]);
    }

    @Benchmark
    public byte[] snappySimplyCompress_1M() {
        return snappySimplyCompress(binaries[1]);
    }

    @Benchmark
    public byte[] snappySimplyCompress_100M() {
        return snappySimplyCompress(binaries[2]);
    }

    @Benchmark
    public byte[] snappySimplyDecompress_1K() {
        return snappySimplyDecompress(snappySimplyCompressed[0]);
    }

    @Benchmark
    public byte[] snappySimplyDecompress_1M() {
        return snappySimplyDecompress(snappySimplyCompressed[1]);
    }

    @Benchmark
    public byte[] snappySimplyDecompress_100M() {
        return snappySimplyDecompress(snappySimplyCompressed[2]);
    }

    ///// Snappy-Java Stream
    @Benchmark
    public byte[] snappyStreamCompress_1K() {
        return streamBasedCompress(binaries[0], is -> new SnappyOutputStream(is));
    }

    @Benchmark
    public byte[] snappyStreamCompress_1M() {
        return streamBasedCompress(binaries[1], is -> new SnappyOutputStream(is));
    }

    @Benchmark
    public byte[] snappyStreamCompress_100M() {
        return streamBasedCompress(binaries[2], is -> new SnappyOutputStream(is));
    }

    @Benchmark
    public byte[] snappyStreamDecompress_1K() {
        return streamBasedDecompress(snappyStreamCompressed[0], is -> new SnappyInputStream(is));
    }

    @Benchmark
    public byte[] snappyStreamDecompress_1M() {
        return streamBasedDecompress(snappyStreamCompressed[1], is -> new SnappyInputStream(is));
    }

    @Benchmark
    public byte[] snappyStreamDecompress_100M() {
        return streamBasedDecompress(snappyStreamCompressed[2], is -> new SnappyInputStream(is));
    }

    ///// LZ4-Java
    @Benchmark
    public byte[] lz4Compress_1K() {
        return lz4Compress(binaries[0]);
    }

    @Benchmark
    public byte[] lz4Compress_1M() {
        return lz4Compress(binaries[1]);
    }

    @Benchmark
    public byte[] lz4Compress_100M() {
        return lz4Compress(binaries[2]);
    }

    @Benchmark
    public byte[] lz4Decompress_1K() {
        return lz4Decompress(lz4Compressed[0], binaries[0].length);
    }

    @Benchmark
    public byte[] lz4Decompress_1M() {
        return lz4Decompress(lz4Compressed[1], binaries[1].length);
    }

    @Benchmark
    public byte[] lz4Decompress_100M() {
        return lz4Decompress(lz4Compressed[2], binaries[2].length);
    }

    @FunctionalInterface
    interface IoFunction<I, O> {
        O apply(I i) throws IOException;
    }

    byte[] streamBasedCompress(byte[] binary, IoFunction<OutputStream, OutputStream> fun) {
        ByteArrayOutputStream compressBaos = new ByteArrayOutputStream();
        try (OutputStream compressStream = fun.apply(compressBaos)) {
            compressStream.write(binary);
        } catch (IOException e) {
            throw new UncheckedIOException(e);
        }

        return compressBaos.toByteArray();
    }

    byte[] streamBasedDecompress(byte[] compressed, IoFunction<InputStream, InputStream> fun) {
        ByteArrayOutputStream decompressBaos = new ByteArrayOutputStream();
        try (InputStream decompressStream = fun.apply(new ByteArrayInputStream(compressed))) {
            int b;
            while ((b = decompressStream.read()) != -1) {
                decompressBaos.write(b);
            }
        } catch (IOException e) {
            throw new UncheckedIOException(e);
        }

        return decompressBaos.toByteArray();
    }

    byte[] snappySimplyCompress(byte[] binary) {
        try {
            return Snappy.compress(binary);
        } catch (IOException e) {
            throw new UncheckedIOException(e);
        }
    }

    byte[] snappySimplyDecompress(byte[] compressed) {
        try {
            return Snappy.uncompress(compressed);
        } catch (IOException e) {
            throw new UncheckedIOException(e);
        }
    }

    byte[] lz4Compress(byte[] binary) {
        LZ4Factory factory = LZ4Factory.fastestInstance();
        LZ4Compressor compressor = factory.fastCompressor();

        return compressor.compress(binary);
    }

    byte[] lz4Decompress(byte[] compressed, int originalSize) {
        LZ4Factory factory = LZ4Factory.fastestInstance();
        LZ4FastDecompressor decompressor = factory.fastDecompressor();

        return decompressor.decompress(compressed, originalSize);
    }
}