Filesystem Compression: Transparent Space Savings and Performance Trade-offs

What Is Transparent Compression?

Transparent compression = Automatic compression/decompression by the filesystem—invisible to applications.

How it works:

Application writes uncompressed data
Filesystem compresses before writing to disk
Application reads file
Filesystem decompresses on-the-fly
Application never knows compression happened

Benefits:

Space savings: 30-70% less disk usage (depends on data)
I/O performance: Read/write less data to disk
Cost reduction: Smaller storage needs
Bandwidth: Faster over slow disks (CPU decompress faster than disk read)

Trade-offs:

CPU overhead: Compression/decompression uses CPU
Incompressible data: Images/videos get no benefit (already compressed)
Memory usage: Decompression buffers

How Compression Works: Interactive Exploration

See transparent compression in action—from write to read, with different algorithms:

Interactive Compression Demo

Transparent Compression: Write and Read Flow

Step 1: Application Writes Uncompressed Data

Application Data:

Size:100KB

[2025-01-10] Log entry 1... [2025-01-10] Log entry 2...

What's happening:

→Application writes 100KB log data
→Data arrives at filesystem layer uncompressed
→Application unaware of compression (transparent)
→Filesystem will compress before disk write
→No application changes needed!

Step 1 of 5

Compression Algorithms

Speed vs Ratio Trade-off

Fast (Low CPU)          Balanced              Slow (High CPU)
LZ4 → ZSTD (level 1-3) → ZSTD (level 4-9) → ZLIB → LZMA
│                        │                    │
Low ratio               Medium ratio         High ratio
10-30% savings          30-50% savings       40-70% savings

LZ4 (Fastest)

Characteristics:

Speed: Extremely fast (500+ MB/s compression)
Ratio: Low (10-30% savings)
CPU: Minimal overhead
Use case: SSDs, fast storage, CPU-limited systems

Btrfs:

mount -o compress=lz4 /dev/sda1 /mnt

ZFS:

zfs set compression=lz4 tank/data

Performance:

Compression:   500-700 MB/s
Decompression: 1000+ MB/s
Ratio:         ~25% (text), ~10% (binaries)

ZSTD (Balanced - Recommended)

Characteristics:

Speed: Very fast (200-400 MB/s)
Ratio: Good (30-50% savings)
CPU: Low-medium overhead
Levels: 1-22 (default: 3)
Use case: General purpose (best default)

Btrfs:

mount -o compress=zstd /dev/sda1 /mnt
# Or specific level
mount -o compress=zstd:3 /dev/sda1 /mnt

ZFS:

zfs set compression=zstd tank/data
# ZFS uses fixed level, no tuning

Performance by level:

Level 1:  400 MB/s, 30% ratio
Level 3:  250 MB/s, 40% ratio (default)
Level 9:  100 MB/s, 50% ratio
Level 15:  30 MB/s, 55% ratio

ZLIB (High Ratio)

Characteristics:

Speed: Moderate (50-100 MB/s)
Ratio: High (40-60% savings)
CPU: Medium overhead
Use case: Archival, cold storage

Btrfs:

mount -o compress=zlib /dev/sda1 /mnt
# Levels 1-9
mount -o compress=zlib:9 /dev/sda1 /mnt  # Max compression

ZFS:

zfs set compression=gzip tank/archive  # gzip = zlib
# Levels: gzip-1 to gzip-9
zfs set compression=gzip-9 tank/archive

Performance:

Compression:   80-120 MB/s (level 6)
Decompression: 200-300 MB/s
Ratio:         ~45% (text), ~35% (binaries)

LZMA/XZ (Highest Ratio)

Characteristics:

Speed: Slow (10-30 MB/s)
Ratio: Very high (50-70% savings)
CPU: High overhead
Use case: Archival only (too slow for active use)

Not recommended for filesystem compression (too slow)

Algorithm Comparison

Algorithm	Speed	Ratio	CPU	Best For
LZ4	⚡⚡⚡⚡⚡	⭐⭐	Low	SSD, real-time
ZSTD-1	⚡⚡⚡⚡	⭐⭐⭐	Low	General purpose
ZSTD-3	⚡⚡⚡	⭐⭐⭐⭐	Med	Default (best balance)
ZLIB-6	⚡⚡	⭐⭐⭐⭐⭐	Med-High	Cold storage
GZIP-9	⚡	⭐⭐⭐⭐⭐	High	Archival

Filesystem-Specific Compression

Btrfs Compression

Enable at mount:

# Algorithm only (default level)
mount -o compress=zstd /dev/sda1 /mnt

# Algorithm with level
mount -o compress=zstd:3 /dev/sda1 /mnt

# Force compress (even if ratio poor)
mount -o compress-force=zstd /dev/sda1 /mnt

Heuristics:

# compress (default): Skip if ratio <80%
mount -o compress=zstd /dev/sda1 /mnt

# compress-force: Always compress (even if grows)
mount -o compress-force=zstd /dev/sda1 /mnt

Per-file compression:

# Set compression for file/directory
btrfs property set /path/to/file compression zstd

# Disable for specific files
btrfs property set /var/lib/mysql/data compression none

Defragment with compression:

# Recompress existing files
btrfs filesystem defragment -r -czstd /mnt

ZFS Compression

Enable per dataset:

# Algorithm
zfs set compression=zstd tank/data

# Compression levels (gzip only)
zfs set compression=gzip-9 tank/archive  # Max

# Disable
zfs set compression=off tank/raw

Inheritance:

# Set on parent, children inherit
zfs set compression=zstd tank
# All datasets under tank/ use zstd

# Override on child
zfs set compression=lz4 tank/fast

Check compression ratio:

zfs get compressratio tank/data
# NAME        PROPERTY       VALUE
# tank/data   compressratio  2.13x  (53% savings)

Compression + dedup:

# Can combine (but expensive)
zfs set compression=zstd tank/data
zfs set dedup=on tank/data  # Use with caution (RAM heavy)

NTFS Compression (Windows)

Enable on Windows:

# Right-click → Properties → Advanced → Compress
# Or via command:
compact /c /s:C:\path\to\compress

Algorithm:

Fixed algorithm (LZNT1 - similar to LZ77)
No tuning options
Per-file or per-directory

Linux NTFS-3G:

mount -t ntfs-3g -o compression /dev/sda1 /mnt

Performance:

Slower than modern algorithms
~30-40% ratio on text
Not recommended for active use

ReFS Compression (Windows Server)

New in ReFS:

# Per-file compression (Server 2022+)
Set-ItemProperty -Path C:\file.txt -Name Compression -Value LZ4

Algorithms:

LZ4 (fast)
Uncompressed (default)

When Compression Helps

Compressible Data

High compression (50-70%):

Text files (.txt, .log, .csv)
Source code (.c, .py, .js)
Configuration files (.conf, .xml, .json)
Database dumps (.sql)
Uncompressed images (.bmp, .tiff)

Medium compression (30-50%):

Binary executables (some patterns)
Virtual machine images (depends on content)
Documents (.doc, .pdf - if not already compressed)

Low/no compression (0-10%):

Compressed files (.zip, .gz, .xz)
Images (.jpg, .png, .webp)
Videos (.mp4, .mkv, .webm)
Audio (.mp3, .flac, .opus)
Encrypted data (appears random)

Performance Scenarios

Compression HELPS performance when:

Disk bandwidth < (CPU decompression speed)

Example:
HDD:     100 MB/s read
ZSTD:    400 MB/s decompress
2x ratio: Effective 200 MB/s (2x faster!)

Compression HURTS performance when:

Disk bandwidth > (CPU decompression speed)

Example:
NVMe SSD: 3000 MB/s read
ZSTD:     400 MB/s decompress
Result:   Bottlenecked by CPU (slower!)

Compression Ratios: Real World

Text Data

Source code (Linux kernel):
  Original:    1000 MB
  LZ4:          750 MB (25% savings, 1.33x)
  ZSTD-3:       600 MB (40% savings, 1.67x)
  ZLIB-6:       550 MB (45% savings, 1.82x)

Database Data

PostgreSQL database dump:
  Original:    10 GB
  LZ4:          7 GB (30% savings, 1.43x)
  ZSTD-3:       5 GB (50% savings, 2.00x)
  ZLIB-6:       4 GB (60% savings, 2.50x)

Mixed Filesystem

Typical Linux root filesystem:
  Original:    20 GB
  LZ4:         17 GB (15% savings, 1.18x)
  ZSTD-3:      14 GB (30% savings, 1.43x)
  ZLIB-6:      12 GB (40% savings, 1.67x)

Lower than pure text (includes binaries, libs)

Performance Impact

CPU Overhead

Compression (write):

LZ4:     5-10% CPU
ZSTD-1:  10-15% CPU
ZSTD-3:  15-25% CPU
ZLIB-6:  30-50% CPU

Decompression (read):

LZ4:     2-5% CPU
ZSTD:    5-10% CPU
ZLIB:    10-20% CPU

Modern CPUs: Overhead often negligible (multi-core, idle cores)

Throughput

Sequential read (1GB file):

No compression:
  HDD:  100 MB/s
  SSD:  500 MB/s

With ZSTD (2x ratio):
  HDD:  200 MB/s effective (2x improvement!)
  SSD:  400 MB/s effective (CPU bottleneck)

Random I/O:

Compression helps less (can't skip decompression)
Smaller chunks = more overhead

Best Practices

1. Choose Right Algorithm

General purpose:

# Btrfs
mount -o compress=zstd /dev/sda1 /mnt

# ZFS
zfs set compression=zstd tank/data

Fast SSD:

# Use LZ4 or disable
mount -o compress=lz4 /dev/nvme0n1 /mnt
# Or no compression (SSD fast enough)

Archival/cold storage:

# Use high compression
mount -o compress=zlib:9 /dev/sda1 /archive
zfs set compression=gzip-9 tank/archive

2. Selective Compression

Enable for compressible data:

# Btrfs: Compress /home, logs
btrfs property set /home compression zstd
btrfs property set /var/log compression zstd

# Don't compress videos, photos
btrfs property set /media compression none

ZFS datasets:

zfs create -o compression=zstd tank/code    # Source code
zfs create -o compression=lz4 tank/vm       # VM images
zfs create -o compression=off tank/media    # Videos/photos

3. Monitor Compression Ratio

Btrfs:

compsize /mnt
# Output:
# Processed 12345 files, 10GB
# Compressed size: 6GB
# Ratio: 1.67x (40% savings)

ZFS:

zfs get compressratio,used,referenced tank/data
# compressratio  2.13x
# used          47GB
# referenced    100GB  (100GB → 47GB on disk)

4. Existing Data

Btrfs recompress:

# Defragment with compression
btrfs filesystem defragment -r -czstd /mnt

# Or per-file
find /mnt -type f -exec btrfs filesystem defragment -czstd {} \;

ZFS: Data compressed on next write only

# Enable compression
zfs set compression=zstd tank/data

# Rewrite data to compress
# (no built-in recompress - must copy/rewrite)

5. Database Considerations

Don't compress database files:

# Databases have own compression
# Filesystem compression = double CPU overhead

# Btrfs: Disable for databases
chattr +C /var/lib/mysql  # Also disables CoW
btrfs property set /var/lib/mysql compression none

# ZFS: Disable
zfs set compression=off tank/mysql

Limitations

What Compression Can't Do

Incompressible data: Can't compress JPEG, MP4, ZIP (already compressed)
Random writes: May increase fragmentation
CPU bound: Slow CPUs = performance loss
Memory: Decompression needs RAM buffers

Compression + Other Features

Btrfs:

✅ Compression + CoW: Works well
✅ Compression + Snapshots: Shares compressed blocks
⚠️ Compression + Defrag: Breaks sharing (recompresses)

ZFS:

✅ Compression + CoW: Perfect together
✅ Compression + Snapshots: Deduplicates compressed
⚠️ Compression + Dedup: High memory usage
✅ Compression + Encryption: Compress before encrypt (better)

Advanced: Compression Order

ZFS Compression + Encryption

Order matters:

Write path:
1. Application data (plain)
2. Compress (reduces size)
3. Encrypt (encrypted compressed data)
4. Write to disk

Why: Compression needs patterns (encryption destroys patterns)

Enable both:

zfs create -o compression=zstd -o encryption=aes-256-gcm tank/secure
# Automatically: compress then encrypt

Deduplication + Compression

Can combine but expensive:

# ZFS: Both enabled
zfs set compression=zstd tank/data
zfs set dedup=on tank/data

# Dedup on compressed blocks
# High RAM usage (256 bytes per block in ARC)

Recommendation: Use compression OR dedup, rarely both

Troubleshooting

Check If Compression Active

Btrfs:

# Check mount options
mount | grep compress

# Check per-file
btrfs property get /path/to/file compression

# Measure ratio
compsize /mnt

ZFS:

# Check dataset property
zfs get compression tank/data

# Check actual ratio
zfs get compressratio tank/data

Poor Compression Ratio

Causes:

Incompressible data (JPEG, MP4, ZIP)
Already compressed files
Encrypted data
Random/binary data

Solution:

# Disable for incompressible paths
btrfs property set /media compression none
zfs set compression=off tank/media

Performance Degradation

If compression slows system:

# Switch to faster algorithm
mount -o remount,compress=lz4 /mnt
zfs set compression=lz4 tank/data

# Or disable entirely
mount -o remount,compress=no /mnt
zfs set compression=off tank/data

Copy-on-Write: Works seamlessly with compression
Snapshots: Share compressed blocks
Btrfs: Transparent compression support
ZFS: Compression with encryption
Mount Options: Configuring compression

Key Takeaways

Transparent: Automatic compression/decompression (invisible to apps)
Algorithm: ZSTD recommended (best balance speed/ratio)
Space Savings: 30-50% typical (depends on data type)
Performance: Can improve I/O on slow disks (HDD)
CPU Overhead: Minimal on modern CPUs (5-15%)
Selective: Compress text/code, skip media/videos
Incompressible: JPEG, MP4, ZIP get no benefit
Enable Now: Modern filesystems (Btrfs, ZFS, ReFS)