Latest Benchmark Results

This page shows detailed benchmark results. Results are updated with each release.

Memory Scaling

Memory usage (RSS) when watching multiple SQLite databases:

Databases	walrust (MB)	litestream (MB)	Reduction
1	19	36	47%
10	19	55	65%
100	20	160	88%

Measured with 100KB databases, syncing to Tigris S3 on macOS.

Startup Time

Time to start watching N databases (mean of 3 runs):

Databases	Startup Time (ms)
1	45
10	82
100	423

Change Detection Latency

Time from SQLite write to walrust detection:

Databases	p50 (ms)	p95 (ms)	p99 (ms)
1	2.1	4.8	8.2
10	4.3	12.1	18.7
100	8.7	28.4	52.3

CPU Usage Under Load

Average CPU usage during concurrent writes:

Databases	CPU %
1	2.1
10	8.4
100	24.7

Sync Latency (End-to-End)

Time from SQLite write to data available in S3:

Operation	p50 (ms)	p95 (ms)	p99 (ms)
Single write	45	82	124
Batch (100)	52	98	156

Note: Using MinIO local storage. Real S3 latency will be higher due to network.

Restore Performance

Time to restore database from S3:

Database Size	Restore Time (ms)
100 KB	234
1 MB	412
10 MB	1,847

Test Environment

These results were collected on:

• Platform: macOS (Apple Silicon)
• Storage: Tigris S3
• Database size: 100KB each
• walrust: v0.3.0
• litestream: v0.5.2

Running Your Own Benchmarks

# Start MinIO
make bench-minio

# Run and get JSON output
python bench/compare.py --use-minio --json
python bench/multidb.py --use-minio --json
python bench/realworld.py --use-minio --json

CI Artifacts

Benchmark JSON files are available as artifacts on each release:

• Latest release artifacts