Skip to main content
XZ is a modern compression format using the LZMA2 algorithm, providing excellent compression ratios with reasonable decompression speed.

​
Format Overview

XZ provides:
  • Excellent compression ratio - Best general-purpose compression
  • LZMA2 algorithm - Advanced dictionary-based compression
  • Block structure - Independent blocks for parallelization
  • Integrity checking - CRC32, CRC64, SHA-256 support
  • No size limits - Files up to 16 exabytes
  • Stream padding - Multiple concatenated streams
XZ typically achieves 10-30% better compression than BZIP2 and 20-50% better than GZIP, making it ideal for distribution and archival.

​
Format Structure

From source/CPP/7zip/Archive/XzHandler.cpp:37-43, XZ uses a block-based structure: 
struct CBlockInfo {
  unsigned StreamFlags;
  UInt64 PackPos;
  UInt64 PackSize;    // Size from Index record (no padding)
  UInt64 UnpackPos;
};

​
Stream Format

+-------------------------+
| Stream Header (12 bytes)|
+-------------------------+
| Block(s)                |
| - Block Header          |
| - Compressed Data       |
| - Block Padding         |
| - Check Value           |
+-------------------------+
| Index                   |
+-------------------------+
| Stream Footer (12 bytes)|
+-------------------------+

​
Stream Header

Bytes 0-5:   Magic (0xFD '7' 'z' 'X' 'Z' 0x00)
Byte 6-7:    Stream Flags
Bytes 8-11:  CRC32 of bytes 6-7
The XZ magic bytes FD 37 7A 58 5A 00 make XZ files easily identifiable and less likely to be confused with other formats.

​
Compression Algorithm

​
LZMA2 Method

From XzHandler.cpp:34: 
#define k_LZMA2_Name "LZMA2"
LZMA2 features:
  • Dictionary matching - Finds repeated patterns
  • Range encoding - Efficient entropy coding
  • Multi-threading support - Parallel compression
  • Better than LZMA - Reset states, better for incompressible data

​
Filters

XZ supports preprocessing filters (XzHandler.cpp:101): 
UInt32 _filterId;  // BCJ, Delta, etc.
Common filters:
  • BCJ (x86) - For x86 executables
  • BCJ (ARM, ARM64, PPC, IA64, SPARC) - For other architectures
  • Delta - For numeric data

​
Usage Examples

​
Compress Single File

7z a -txz file.txt.xz file.txt

​
Compress with Maximum Compression

7z a -txz -mx=9 file.xz file.txt

​
Compress with Specific Dictionary Size

7z a -txz -mx=9 -md=64m file.xz file.txt
Dictionary sizes:
  • -md=16m - Fast, moderate compression
  • -md=32m - Good balance (default for -mx=9)
  • -md=64m - Better compression, more memory
  • -md=128m - Maximum compression, high memory

​
Decompress File

7z x file.xz

​
Create TAR.XZ Archive

7z a -ttar archive.tar files/
7z a -txz -mx=9 archive.tar.xz archive.tar
Or using system tar: 
tar -cJf archive.tar.xz files/

​
Multi-threaded Compression

7z a -txz -mx=9 -mmt=4 file.xz large_file.txt
Uses 4 threads for faster compression.
Multi-threaded XZ compression can dramatically reduce compression time on multi-core systems, though it may slightly reduce compression ratio.

​
Handler Implementation

From source/CPP/7zip/Archive/XzHandler.cpp:46-78: 
Z7_class_CHandler_final:
  public IInArchive,
  public IArchiveOpenSeq,
  public IInArchiveGetStream,
  public ISetProperties,
 #ifndef Z7_EXTRACT_ONLY
  public IOutArchive,
 #endif
  public CMyUnknownImp,
 #ifndef Z7_EXTRACT_ONLY
  public CMultiMethodProps
 #else
  public CCommonMethodProps
 #endif

​
Handler State

bool _stat_defined;
bool _stat2_defined;
bool _isArc;
bool _needSeekToStart;
bool _firstBlockWasRead;
SRes _stat2_decode_SRes;

CXzStatInfo _stat;    // Backward parsing statistics
CXzStatInfo _stat2;   // Forward parsing statistics

​
Archive Properties

From XzHandler.cpp:196-200: 
static const Byte kProps[] = {
  kpidSize,
  kpidPackSize,
  kpidMethod
};
XZ tracks:
  • Size - Uncompressed size
  • PackSize - Compressed size
  • Method - Compression method (LZMA2)
  • NumStreams - Number of streams
  • NumBlocks - Number of blocks

​
Compression Levels

​
Performance Comparison (100 MB mixed files)

LevelDict SizeTimeSizeRatioMemory (Comp)Memory (Decomp)
-mx=0-5s80 MB80%3 MB1 MB
-mx=1256 KB20s35 MB35%10 MB2 MB
-mx=34 MB35s28 MB28%50 MB6 MB
-mx=58 MB45s25 MB25%100 MB10 MB
-mx=716 MB60s23 MB23%200 MB18 MB
-mx=964 MB90s20 MB20%700 MB66 MB
Higher compression levels require significantly more memory. Ensure your system has sufficient RAM before using -mx=9 with large dictionary sizes.

​
Advanced Features

​
Block Size Configuration

UInt64 _numSolidBytes;  // Block size (XzHandler.cpp:107)
Set block size: 
7z a -txz -mx=9 -ms=64m file.xz large_file.txt
Block sizes:
  • Smaller blocks - Better for random access, multi-threading
  • Larger blocks - Better compression ratio

​
Integrity Checks

XZ supports multiple check types:
  • None - No integrity check (fastest)
  • CRC32 - Fast, good detection
  • CRC64 - Better detection
  • SHA-256 - Cryptographic hash
7z a -txz -mx=9 -mcheck=crc64 file.xz file.txt

​
Stream Concatenation

Multiple XZ streams can be concatenated: 
7z a -txz file1.xz file1.txt
7z a -txz file2.xz file2.txt
cat file1.xz file2.xz > combined.xz
7z x combined.xz  # Extracts both streams

​
Filters for Executables

Use BCJ filter for better executable compression: 
7z a -txz -mx=9 -mf=BCJ program.xz program.exe
Filter types:
  • BCJ - x86 executables
  • ARM - ARM binaries
  • ARM64 - ARM64 binaries
  • PPC - PowerPC binaries

​
Memory Requirements

​
Compression Memory

Memory ≈ Dictionary × 11.5 + 100 MB
Examples:
  • -md=8m: ~200 MB
  • -md=32m: ~470 MB
  • -md=64m: ~840 MB
  • -md=128m: ~1.6 GB

​
Decompression Memory

Memory ≈ Dictionary + 10 MB
Examples:
  • 8 MB dict: ~18 MB
  • 32 MB dict: ~42 MB
  • 64 MB dict: ~74 MB
Decompression is much more memory-efficient than compression, allowing archives compressed with -mx=9 to be extracted on low-memory systems.

​
Implementation Details

XZ implementation files: 
source/CPP/7zip/Archive/XzHandler.cpp    # Archive handler
source/CPP/7zip/Compress/XzDecoder.cpp   # Decompression
source/CPP/7zip/Compress/XzEncoder.cpp   # Compression
Includes C library: 
#include "../../../C/Alloc.h"
Uses 7-Zip’s XZ implementation compatible with XZ Utils.

​
Comparison with Other Formats

​
XZ vs GZIP

FeatureXZGZIP
Compression ratioExcellentGood
Compression speedSlowFast
Decompression speedMediumFast
Memory usageHighLow
Best forDistribution, archivalQuick compression

​
XZ vs BZIP2

FeatureXZBZIP2
Compression ratio15-20% betterGood
Compression speedSimilarSimilar
Decompression speedFasterSlower
Memory usageHigherLower
Block independenceYesYes

​
XZ vs 7z

FeatureXZ7z
Multiple filesNo (needs TAR)Yes
Solid compressionBlock-basedFull solid support
EncryptionNoYes (AES-256)
Compression ratioExcellentExcellent (similar)
CompatibilityGrowingRequires 7-Zip

​
Best Practices

For Distribution

Use -mx=9 with moderate dictionary for best compression

For Large Files

Use multi-threading: -mmt=4 or more

For Archives

Combine with TAR: .tar.xz format

For Speed

Use -mx=3 to -mx=5 for faster compression

​
Common Use Cases

​
Software Distribution

tar -cJf myapp-1.0.tar.xz myapp-1.0/
XZ is becoming the standard for Linux source distributions.

​
System Backups

tar -cJf backup-$(date +%Y%m%d).tar.xz /home /etc

​
Log Archival

find /var/log -name "*.log" -mtime +30 -exec xz {} \;

​
Database Dumps

mysqldump --all-databases | 7z a -txz -mx=9 -si backup.sql.xz

​
Performance Optimization

​
Parallel Compression

Use multi-threading: 
7z a -txz -mx=9 -mmt=on file.xz large_file.txt
Or specify thread count: 
7z a -txz -mx=9 -mmt=8 file.xz large_file.txt

​
Dictionary Size Selection

Choose based on file size and available RAM:
  • File < 1 MB - Use -md=1m
  • File 1-10 MB - Use -md=8m
  • File 10-100 MB - Use -md=32m
  • File > 100 MB - Use -md=64m or higher

​
Preset Levels

7z a -txz -mx=6 file.xz data.txt  # Good balance
Recommended presets:
  • -mx=3 - Fast, good compression
  • -mx=6 - Balanced (similar to xz -6)
  • -mx=9 - Maximum compression

​
Limitations

XZ format limitations:
  • Single file compression (use TAR for multiple files)
  • High memory requirements for compression
  • Slower than GZIP and BZIP2 for compression
  • No encryption support
  • Requires modern tools (not on very old systems)

​
Compatibility

​
Tool Support

  • xz-utils - Reference implementation (Linux)
  • 7-Zip - Full support (Windows, Linux)
  • XZ Utils - Native support (macOS via Homebrew)
  • PeaZip - GUI support
  • File Roller, Ark - Linux GUI support

​
Platform Support

  • Linux - Universal (xz command)
  • macOS - Via Homebrew: brew install xz
  • Windows - Via 7-Zip
  • BSD - Built-in or via packages

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us