Kolmos Logo
Kolmos
Home/Journal/Benchmarks
Benchmarks·10 min read·November 1, 2026

KOLMOS Benchmarks: Real-World Latency, Ingest Throughput & Storage Compression Ratios

Empirical benchmark evaluation of KOLMOS on TPC-H Lineitem, NYC Taxi Trips, User Clickstream, and MongoDB BSON datasets measured against Apache Parquet-zstd and physical cloud object storage.

KOLMOS Engineering Team
KOLMOS Engineering Team
Benchmarking & Performance Systems · KOLMOS Systems
Follow on LinkedIn
KOLMOS Benchmarks: Real-World Performance and Storage
KOLMOS Benchmarks: Real-World Performance and Storage
Methodology Standard: All benchmark figures published by KOLMOS are gathered on physical release builds (--release) running on standard cloud hardware. We test against Apache Parquet with zstd level 3 compression as the industry gold standard for columnar storage.

Summary of Measured Benchmark Results

  • Compression Advantage: 1.66× to 2.12× smaller physical byte size than Apache Parquet-zstd.
  • Ingestion Throughput: 194,000 to 329,000 rows/second per single CPU core.
  • Point Lookup Latency: Sub-millisecond to 4ms via in-memory PK Bloom filter sidecars.
  • Lossless Recovery: 100% bit-exact decode fidelity verified with BLAKE3 cryptographic hashes.
    • Technical Blueprint
    Storage PlaneCloudflare R2 CAS
    Execution CoreDataFusion SIMD
    Decode Fidelity100% Bit-Exact
    Wire DoorsPG · MySQL · Mongo
    Official Channels
    linkedin.com/company/kolmos
    kolmos.dev
    Related Articles
    PostgreSQL
    PostgreSQL Compression: How to Reduce Database Storage Without Sacrificing Performance
    PostgreSQL
    How to Make PostgreSQL Cheaper Without Sacrificing Performance
    Cloud Costs
    How to Reduce Amazon Aurora Costs in 2026: Taming the I/O and Storage Beast

    Dataset Deep Dives

    1. TPC-H Lineitem (500,000 Relational Rows)

  • Raw CSV Size: 59.2 MB
  • Apache Parquet-zstd Size: 12.7 MB
  • KOLMOS Size (.ksf): 7.67 MB (1.656× smaller than Parquet)
  • Single-Core Ingest Speed: 248,000 rows/second
  • Characteristics: Standard transactional schema with 16 columns including dates, quantities, decimal prices, and status flags.

    • 2. NYC Yellow Taxi Trips (200,000 Time-Series Rows)

  • Raw CSV Size: 28.4 MB
  • Apache Parquet-zstd Size: 3.50 MB
  • KOLMOS Size (.ksf): 1.61 MB (1.892× smaller than Parquet)
  • Single-Core Ingest Speed: 329,000 rows/second
  • Compression Mechanism: Autonomous formula mining automatically discovered fare relationships (total_amount = fare + tip + tolls), storing exact derivations with residual deltas.

    • 3. User Clickstream Telemetry (500,000 Log Events)

  • Raw JSON Size: 68.5 MB
  • Apache Parquet-zstd Size: 5.16 MB
  • KOLMOS Size (.ksf): 1.89 MB (2.117× smaller than Parquet)
  • Single-Core Ingest Speed: 194,000 rows/second
  • Compression Mechanism: Prototype clustering grouped recurring user session patterns, encoding rows as delta offsets from cluster medoids.

    • 4. Live MongoDB BSON Documents (1,000,000 Docs on Cloudflare R2)

  • Raw Uncompressed JSON: 71.3 MB
  • KOLMOS Stored in Cloudflare R2: 2.60 MB (27.7× vs Raw, 2.08× vs Parquet)
  • Live Scan Query Time: 24.0 ms over live R2 HTTPS fetch.

    • Next Steps & Interactive Tools

  • Interactive Database Benchmarks Page — Use the live dataset inspector and interactive cloud cost calculator.
  • What is KOLMOS? — Learn about the multi-wire unified database architecture.
  • KOLMOS vs PostgreSQL — Compare storage and query performance.
  • Official Homepage — Discover the cloud database engine.
    • Try KOLMOS Today

    Deploy Your First Self-Compressing Store.

    Connect via PostgreSQL, MySQL, or MongoDB. 10 GB free developer storage included.

    Start Free Trial ⚡Inspect Architecture