Empirical numbers,
zero marketing fluff.

Every metric below was measured on physical release builds against Apache Parquet-zstd and Cloudflare R2 object storage. No unmeasured claims.

Inspect Measured Datasets Cloud Cost Calculator

1.66× – 2.12×

Smaller than Parquet-zstd

Measured on real physical datasets

194K – 329K

Rows Ingested per Second

Single-core throughput baseline

15× – 40×

Cloud Storage Cost Savings

Cloudflare R2 vs AWS RDS / Atlas

100%

Bit-Exact Decode Fidelity

Verified across all 18 golden fixtures

Measured Datasets

Real Workload Benchmarks.

Select a real dataset below to inspect its physical compression ratio against Apache Parquet-zstd and ingest throughput.

Structured RelationalRung 0 Literal Baseline

TPC-H Lineitem (500K Rows)

Standard relational analytical benchmark with 16 columns (dates, decimals, strings). Demonstrates predictable dictionary-trained compression on traditional warehouse data.

Ingest Throughput

248,000 rows/sec

Total Record Count

500,000 Rows

Storage Footprint Comparison (200K-Row Segments)

Apache Parquet-zstd (Baseline)12.7 MB

KOLMOS Explanation Engine7.67 MB

Physical Compression Advantage

1.656× smaller

Cost Economics

Real Cloud Storage Savings.

Comparing AWS RDS gp3 ($0.115/GB/mo) and MongoDB Atlas ($0.25/GB/mo) against KOLMOS on Cloudflare R2 ($0.015/GB/mo with 2× physical compression).

AWS RDS (PostgreSQL / MySQL)

$11,500/mo

gp3 EBS volume storage at standard AWS rates ($0.115 per GB-month).

MongoDB Atlas Cluster

$25,000/mo

Managed Atlas dedicated tier storage & replica snapshot IOPS.

15×–33× Cheaper

KOLMOS on Cloudflare R2

$750/mo

100TB compressed to ~50TB on Cloudflare R2 ($0.015 per GB-month).

* Zero egress fees on Cloudflare R2. Calculations assume conservative 2× physical compression ratio.

The Engineering Tradeoff Explained

Honest Scan Latency vs Storage Economics.

KOLMOS is designed to solve the multi-million dollar storage bloat problem. In exchange for 1.66×–2.12× smaller footprint over Parquet-zstd, brute-force full-table scans are 1.6×–8× slower because mathematical formulas must be decoded.

Point Lookups & Pruned Queries

Sub-Millisecond (~1ms)

Bloom filters and header min/max zone maps skip 99% of segments, ensuring real application queries execute instantly.

Brute-Force Full Table Scans

1.6× – 8× Slower than Parquet

When scanning 100M rows unindexed, Parquet is faster at raw RAM reads; KOLMOS delivers 2× smaller disk storage.

Reproduce Benchmark Results via CLI

# 1. Run local engine benchmark suite on release build
$ cargo bench --bench ladder_bench

# 2. Run Cloudflare R2 live benchmark test
$ powershell -ExecutionPolicy Bypass -File .\run_r2_benchmark.ps1

# 3. Check live R2 storage statistics
$ kolmos --store s3 --s3-bucket kolmos-prod stats

Test KOLMOS on
your datasets.

Point KOLMOS at your existing PostgreSQL, MySQL, or MongoDB database and measure your storage savings in minutes.

Start Free Trial ⚡View Pricing & Plans