Skip to content

Performance Benchmarks

Comparative performance tests of PII Anonymizer and alternatives.

Methodology

What We Measure

  • Anonymization speed -- number of texts processed per second
  • Total time -- time to process the entire test dataset

What We Do NOT Measure

  • Service startup time
  • Initialization/warmup time
  • Memory consumption
  • HTTP API response time (for PII Anonymizer, measured directly via library)

Test Data

  • 1,700 texts with various PII types
  • PII types: email, phone, IP, passport, SNILS, AWS keys, SSH keys
  • Combined texts with multiple PII entries
  • Diverse formats for each type

Testing Conditions

  • PII Anonymizer: directly via library (without HTTP)
  • Presidio: via HTTP API (Docker container)
  • Release builds for all services

Results

PII Anonymizer

Metric Value
Texts processed 1,700
Total time 0.007 sec
Throughput 243,473 texts/sec

Presidio

Metric Value
Texts processed 1,100
Total time 13.692 sec
Throughput 80 texts/sec

Note: Presidio benchmark was run via HTTP API (2 containers: analyzer + anonymizer)

Comparison Table

Service Throughput Relative Speed
PII Anonymizer 243,473 texts/sec 1.0x (baseline)
Presidio 80 texts/sec 0.0003x (3,043x slower)

Visual Comparison

PII Anonymizer: █████████████████████████████████████████████████████████████████████ 243,473 texts/sec
Presidio:       ▏                                                                   80 texts/sec

Technical Details

PII Anonymizer

  • Language: Rust
  • Detection method: Regex patterns
  • Strategy: Mask (default)
  • Patterns: 14 active
  • Runtime: Native binary, release build

Presidio

  • Language: Python
  • Detection method: NLP (spaCy) + Regex
  • Strategy: Replace (default)
  • Runtime: Docker container

Benchmark Structure

benchmarks/
├── benchmark_pii_anonymizer.rs    # PII Anonymizer benchmark (Rust)
├── benchmark_presidio.py          # Presidio benchmark (Python)
├── run_benchmarks.sh              # Main run script
├── scripts/
│   └── presidio_server.sh         # Presidio management script
├── data/                          # Test data (generated automatically)
└── results/                       # Benchmark results
    ├── pii_anonymizer.log
    ├── presidio.log
    └── benchmark_results.txt

Running Benchmarks

Full Benchmark (PII Anonymizer + Presidio)

cd benchmarks
./run_benchmarks.sh

PII Anonymizer Only

cargo run --release --bin benchmark_pii_anonymizer

Presidio Only

# Start Presidio
bash benchmarks/scripts/presidio_server.sh start

# Run benchmark
python3 benchmarks/benchmark_presidio.py http://localhost:5002

# Stop Presidio
bash benchmarks/scripts/presidio_server.sh stop

Notes

  1. Fairness of comparison:
  2. PII Anonymizer is measured directly via library (no HTTP overhead)
  3. Presidio is measured via HTTP API (with network overhead)

  4. This gives Presidio a slight advantage in fairness

  5. Method differences:

  6. PII Anonymizer: regex patterns only
  7. Presidio: NLP model + regex

  8. Presidio is more universal but slower

  9. PII types:

  10. PII Anonymizer: Russia-specific (passport, SNILS, INN) + technical (API keys, SSH)
  11. Presidio: international (email, phone, credit cards, person name)

Conclusions

PII Anonymizer delivers excellent performance thanks to: - Compiled Rust language - Optimized regex patterns - No ML models or NLP analysis - Minimal overhead - 3,043x faster than Presidio!

Presidio provides more intelligent detection: - Contextual analysis with NLP - Name and location recognition - But at a significantly higher computational cost

Key Numbers

  • PII Anonymizer: 243,473 texts/sec (0.007 sec for 1,700 texts)
  • Presidio: 80 texts/sec (13.7 sec for 1,100 texts)
  • Difference: PII Anonymizer is 3,043x faster

The choice depends on your performance requirements and the types of PII you need to detect!

Last updated: April 11, 2026