LZX Compression

Purpose

This guide explains LZX compression, the highest-ratio algorithm used in CAB, CHM, LIT, and OAB formats.

References

• Getting Started Guide

• CAB Format

• CHM Format

• LIT Format

• OAB Format

Concepts

What is LXX?

LZX (Lempel-Ziv Extended) is a high-ratio dictionary compression algorithm with configurable window sizes from 32 KB to 2 MB. It achieves the best compression ratios at the cost of speed.

Key Features

• Configurable window - 15 to 21 bits (32 KB to 2 MB)

• Maximum compression - Best ratio among Microsoft algorithms

• Multiple tables - Huffman tables for different data types

• Block-based - Processes data in blocks for efficiency

Performance characteristics

Metric	Value
Compression ratio	50-70% for text, 40-60% for executables
Compression speed	Slow (1-10 MB/s)
Decompression speed	Medium (10-50 MB/s)
Memory usage	High (32 KB - 2 MB window + overhead)
Window size	Configurable: 2^15 to 2^21 bytes

Window Sizes

Bits	Window Size	Best For
15	32 KB	Small files (<100 KB)
16	64 KB	Medium files (default for CHM)
17	128 KB	Medium-large files
18	256 KB	Large files (1-5 MB)
19	512 KB	Large files (5-10 MB)
20	1 MB	Very large files (10-20 MB)
21	2 MB	Maximum compression, huge files

Usage

CAB files

compressor = Cabriolet::CAB::Compressor.new(compression: :lzx)
compressor.add_file('large-file.dat')
compressor.write('archive.cab')

Command line:

cabriolet create --compression=lzx installer.cab files/

CHM Files with Window Size

compressor = Cabriolet::CHM::Compressor.new

compressor.add_file('index.html', '/index.html', section: :compressed)
compressor.add_file('content.html', '/content.html', section: :compressed)

# Small help file: 15-bit window
compressor.generate('small-help.chm', window_bits: 15)

# Large help file: 21-bit window
compressor.generate('large-help.chm', window_bits: 21)

LIT eBooks

compressor = Cabriolet::LIT::Compressor.new

# HTML compressed with LZX
compressor.add_file('chapter1.html', 'chapter1.html', compress: true)
compressor.add_file('chapter2.html', 'chapter2.html', compress: true)

# Images uncompressed
compressor.add_file('cover.jpg', 'cover.jpg', compress: false)

compressor.generate('book.lit')

OAB with Block Size

compressor = Cabriolet::OAB::Compressor.new

# Default 32 KB blocks
compressor.compress('addressbook.dat', 'full.oab', block_size: 32_768)

# Larger blocks for better compression
compressor.compress('addressbook.dat', 'full.oab', block_size: 65_536)

Compression Effectiveness

Excellent For

• HTML/XML - 65-75% compression

• Text files - 60-70% compression

• Source code - 60-70% compression

• Documentation - 65-75% compression

Good For

• Executables - 50-60% compression

• DLLs - 50-60% compression

• Office documents - 40-50% compression

Poor For

• Already compressed - 0-5% (JPG, PNG, MP3, ZIP)

• Random data - 0-10%

• Encrypted files - 0-5%

• Very small files - Overhead too high

Window Size Selection

Choose window size based on file size:

def choose_window_bits(file_size)
  case file_size
  when 0..50_000
    15  # 32 KB window
  when 50_001..200_000
    16  # 64 KB window
  when 200_001..1_000_000
    17  # 128 KB window
  when 1_000_001..5_000_000
    18  # 256 KB window
  when 5_000_001..20_000_000
    19  # 512 KB window
  else
    21  # 2 MB window
  end
end

file_size = File.size('content.html')
window_bits = choose_window_bits(file_size)

compressor = Cabriolet::CHM::Compressor.new
compressor.generate('help.chm', window_bits: window_bits)

Algorithm Details

Three-Stage Process

1. LZ77 Matching

   • Find repeated sequences in window

   • Encode as (distance, length) pairs

2. Huffman Coding

   • Multiple Huffman tables

   • Separate tables for literals, lengths, distances

3. Aligned Offset Encoding

   • Special encoding for aligned offsets

   • Better compression for structured data

Block Types

• Verbatim - Uncompressed data

• Aligned - LZ77 + Huffman + aligned offsets

• Uncompressed - Raw data (no compression)

Comparison

Algorithm	Ratio	Comp Speed	Decomp Speed	Memory
LZSS	30-50%	Fast	Very Fast	4 KB
MSZIP	40-60%	Medium	Fast	32 KB
LZX	50-70%	Slow	Medium	32KB-2MB
Quantum	45-65%	Medium	Medium	Variable

Best practices

1. Software distribution - Use for installers and downloads

2. Large files - Use larger windows for files >1 MB

3. Small files - Use 15-16 bit windows for <100 KB files

4. eBooks - Excellent for HTML content

5. Skip media - Don’t compress JPG, PNG, MP4

6. Test performance - Verify decompression speed is acceptable

Examples

Create Distribution CAB

require 'cabriolet'

compressor = Cabriolet::CAB::Compressor.new(compression: :lzx)

# Add installer files
compressor.add_file('setup.exe')
compressor.add_file('application.dll')
compressor.add_file('readme.txt')

bytes = compressor.write('installer.cab')

original = Dir['*.{exe,dll,txt}'].sum { |f| File.size(f) }
ratio = (1 - bytes.to_f / original) * 100

puts "LZX compression: #{original} → #{bytes} bytes (#{ratio.round(1)}%)"

Create Help File

require 'cabriolet'

compressor = Cabriolet::CHM::Compressor.new

# Add HTML files (compressed)
Dir.glob('docs/**/*.html').each do |file|
  path = "/#{file.sub('docs/', '')}"
  compressor.add_file(file, path, section: :compressed)
end

# Add images (uncompressed)
Dir.glob('docs/**/*.{png,jpg}').each do |file|
  path = "/#{file.sub('docs/', '')}"
  compressor.add_file(file, path, section: :uncompressed)
end

# Use 18-bit window for large help file
compressor.generate('application-help.chm', window_bits: 18)

Optimize for Size

require 'cabriolet'

# Test different window sizes
results = {}

[15, 16, 18, 21].each do |bits|
  compressor = Cabriolet::CHM::Compressor.new
  compressor.add_file('large-doc.html', '/doc.html', section: :compressed)

  output = "test-#{bits}.chm"
  compressor.generate(output, window_bits: bits)

  results[bits] = File.size(output)
  File.delete(output)
end

# Find best
best = results.min_by { |bits, size| size }
puts "Best window size: #{best[0]} bits (#{best[1]} bytes)"

Troubleshooting

Slow Compression

LZX is inherently slow. To improve:

• Accept slower compression for better ratio

• Use MSZIP if speed critical

• Compress once, decompress many times

High Memory Usage

Large windows use more memory:

• Reduce window size for constrained environments

• Use 15-16 bit windows on low-memory systems

• Consider MSZIP (fixed 32 KB) as alternative

Poor Ratio on Small Files

Window larger than file wastes memory:

• Match window to file size

• Use 15-bit (32 KB) for files <100 KB

• Consider MSZIP for very small files

See also

• MSZIP Compression for faster alternative

• LZSS Compression for lower memory usage

• Choosing a Compression Algorithm

Bibliography

• Wikipedia: LZX

• CHM Specification

• cabextract Documentation