Backup Compression: Balancing Speed, Size, and CPU Usage

Vidensdatabase

Backup compression reduces storage costs and transfer time, but at the cost of CPU usage during backup creation — understanding this trade-off helps you choose appropriate compression settings for your specific situation.

The Core Trade-Off

Higher compression ratios (smaller files) generally require more CPU time and take longer to compress — the right balance depends on your specific priorities: minimizing storage cost, minimizing backup duration, or minimizing CPU impact on a server that's also handling production traffic.

Comparing Common Compression Options

Tool	Compression Ratio	Speed	CPU Usage
gzip	Good, moderate	Fast	Moderate
bzip2	Better than gzip	Slower	Higher
xz	Best compression	Slowest	Highest
zstd	Very good, tunable	Fast, especially at lower levels	Tunable, generally efficient

Using gzip (Common Default)

tar -czvf backup.tar.gz /data

Using zstd (Often the Best Modern Choice)

tar -cf - /data | zstd -o backup.tar.zst

zstd offers a genuinely strong balance — compression ratios competitive with higher-CPU options at much better speed, with tunable compression levels letting you dial in your preferred trade-off point.

Tuning zstd's Compression Level

zstd -3 -o backup.tar.zst backup.tar   # faster, less compression
zstd -19 -o backup.tar.zst backup.tar  # slower, maximum compression

Higher numbers mean better compression at the cost of more time/CPU — test different levels against your actual data to find your preferred balance point.

When to Prioritize Speed Over Compression Ratio

Backing up during active production hours where CPU contention with your application is a real concern
Very large databases where backup duration itself affects your effective RPO
Frequent (e.g. hourly) backups where cumulative CPU time across many backup runs adds up

When to Prioritize Compression Ratio Over Speed

Storage cost is a significant, genuine concern (particularly relevant for long-term retention of many historical backups)
Backups run during genuinely idle periods where CPU usage doesn't compete with anything else
Bandwidth for transferring backups off-server is limited/costly, making smaller files valuable

Considering Multi-Threaded Compression

tar -cf - /data | pigz > backup.tar.gz

pigz is a parallel implementation of gzip, utilizing multiple CPU cores — can significantly speed up compression on a multi-core VPS compared to standard single-threaded gzip, at the same compression ratio.

Testing Compression Options Against Your Actual Data

Compression effectiveness varies significantly based on data type (text compresses very well; already-compressed formats like JPEG/video barely compress further) — test with your actual backup data rather than assuming generic benchmarks apply directly to your specific situation.

Balancing with Your Actual RPO/RTO Needs

See Understanding RPO and RTO: Setting Realistic Recovery Objectives — if backup duration itself threatens your RPO target (backup takes so long that you can't run them frequently enough), prioritize speed even at the cost of larger, less-compressed backup files.

Common Errors

Backup CPU usage causing production performance issues — consider lower compression levels, multi-threaded compression to finish faster, or shifting backup timing to genuinely lower-traffic windows.

Continue Reading

Browse more articles in Backup & Disaster Recovery.

Vidensdatabase

Kategorier

Kategorier