Benchmarks

Sairo ships with a benchmark suite you can run against your own deployment. All numbers on the landing page are derived from these reproducible tests.

Test Environment

Parameter	Value
Runtime	Docker container, single Uvicorn process
Host	macOS, Apple Silicon (Docker Desktop)
Local S3	MinIO (localhost:9000)
Production S3	Remote S3-compatible object storage
Production dataset	134,707 objects, 38.25 TB
Methodology	30 iterations per measurement, percentile reporting

Tip

Production deployments on dedicated Linux hardware with NVMe storage will yield better numbers than these Docker Desktop results.

Search Latency

FTS5 trigram search against the production bucket (134,707 objects, 38.25 TB):

Query	Results	p50	p95
parquet (limit=100)	100	3.1ms	4.8ms
events	200	2.4ms	16.2ms
tracking	200	2.3ms	3.0ms
analytics	200	2.5ms	3.4ms
ingest	200	2.4ms	5.5ms
metadata	200	2.4ms	4.3ms
2026	200	2.2ms	22.0ms

Fastest observed: 1.7ms. Typical queries return in 2-3ms p50 against 134K objects.

Indexing Throughput

Bucket	Objects	Duration	Throughput
bench-small	1,000	1.08s	926 obj/s
bench-mixed	2,416	1.08s	1,348 obj/s
production-bucket	134,707	completed	Production-scale crawl

Indexing rate: 1,000–1,350 objects/second on local MinIO.

Upload Throughput

File Size	p50	Throughput
1 KB	44.9ms	—
1 MB	51.9ms	19.3 MB/s
10 MB	130.7ms	76.5 MB/s
50 MB	436.2ms	114.6 MB/s

API Response Times

Endpoint	p50	p95
/healthz	2.1ms	3.6ms
/api/buckets	4.3ms	5.8ms
Object listing (production)	2.4ms	4.6ms
Presigned URL generation	3.1ms	5.6ms

Concurrent Load

Production S3, concurrent search queries:

Concurrent Users	Requests/sec
5	236
10	333
25	528

Scaling Performance (v3.0)

Benchmarked on production data running in Docker against Leaseweb StorageGRID:

Folder Listing (the user-facing speedup)

Bucket	Objects	Before (v2.0)	After (v3.0)	Speedup
ssp-production-reports	557K	114ms	0.048ms	2,378x
ds-mletl-data	139K	27ms	0.056ms	486x
druid-lw-prod	2M (was >1M, disabled)	312ms (DISTINCT fallback)	0.002ms	191,231x

Folder listing uses pre-computed prefix_children via SQL-only aggregation. Previously disabled for buckets >1M objects due to OOM — now works at any scale.

SQLite PRAGMA Tuning Impact

Query	Before	After	Dataset
COUNT(*)	2.0ms	1.5ms	557K objects
COUNT(*)	11.3ms	7.3ms	2M objects
SUM(size)	40ms	62ms	557K objects
Folder stats rebuild	63ms	56ms	139K objects

PRAGMAs applied: cache_size=-64000 (64MB), mmap_size=268435456 (256MB), temp_store=MEMORY.

Crawl Performance

Setting	Before	After
Crawl workers	6	12
Prefix workers	4	16
Batch size	2,000	10,000
Update chunks	500	2,000
FTS rebuild	Blocks 15+ min	Background thread
Sub-prefix splitting	None	Auto for buckets with few prefixes

At 1PB Scale (Projected)

Operation	Expected Performance
Folder listing	< 1ms (constant time, index lookup)
Search	~1ms (FTS5, always available)
Storage breakdown	~500ms (full scan with PRAGMA tuning)
Crawl (50M objects)	~25 min (with sub-prefix splitting)
FTS rebuild (50M objects)	~83 min (background, non-blocking)

Running the Benchmarks

The benchmark suite lives in the benchmark/ directory of the repository.

1. Seed test data

# Requires MinIO CLI (mc) configured with alias "local"
cd benchmark
./seed-data.sh              # Seeds bench-small (1K) + bench-mixed (2.4K)
./seed-data.sh medium       # Seeds bench-medium (10K objects)
./seed-data.sh large        # Seeds bench-large (50K objects)

The seeder creates four buckets with realistic file structures:

Bucket	Objects	Pattern
bench-small	1,000	5 dirs × 10 months × 20 files
bench-mixed	~2,400	Parquet data lake, logs, configs, CSV reports
bench-medium	10,000	10 dirs × 10 months × 10 sub-dirs × 10 files
bench-large	50,000	10 dirs × 50 partitions × 100 records

2. Run benchmarks

# Run all benchmarks
./run-benchmarks.sh

# Run specific benchmark categories
./run-benchmarks.sh search          # Search latency only
./run-benchmarks.sh crawl           # Crawl/indexing only
./run-benchmarks.sh crawl listing   # Crawl + listing

Prerequisites

• Sairo running on localhost:8000 (or set SAIRO_URL)
• MinIO running on localhost:9000
• Test buckets seeded via seed-data.sh
• Default admin credentials (admin/password) or set ADMIN_USER/ADMIN_PASS

3. Results

Results are saved to benchmark/results/:

• JSON — machine-readable per-run results (benchmark-YYYYMMDD-HHMMSS.json)
• Markdown — human-readable summary (LATEST-RESULTS.md)

Landing Page Claims

Every number on the landing page maps to a specific benchmark result:

Landing Page Claim	Benchmark Evidence
”Single-digit millisecond search”	Production p50 = 2.2–3.1ms on 134K objects
”1,300+ obj/sec indexing”	1,348 obj/s measured on bench-mixed
”Sub-5ms API responses”	healthz p50 = 2.1ms, most endpoints < 5ms
”500+ requests/second”	528 req/s at 25 concurrent users (production)
“114 MB/s upload”	50 MB file upload sustained throughput