Performance you can reproduce.
All benchmarks on this page come with a public test harness. Fork the repo and run them on your own AWS or GCP account.
Methodology note
TPC-DS benchmark data generated at official TPC scales (1TB / 10TB / 50TB). "ETL-first warehouse" baseline uses a representative configuration of copy-to-warehouse pipelines observed in production environments — not a specific vendor. Egress costs calculated at AWS us-east-1 list pricing. All tests run on equivalent compute configurations. Streaming latency measured end-to-end from Kafka produce to lakehouse table read-availability.
SQL query benchmarks at scale
| Query | Description | DataLynxr | ETL-first baseline | Speedup |
|---|---|---|---|---|
| Q47 | Year-over-year sales variance (window fn) | 3.8s | 14.2s | 3.7× |
| Q19 | Brand aggregation with join | 1.4s | 6.9s | 4.9× |
| Q72 | Inventory snapshot with correlated sub-query | 8.1s | 22.6s | 2.8× |
| Q96 | Call center sales join + filter | 0.6s | 3.1s | 5.2× |
10TB TPC-DS scale. 16-core compute node. Parquet + Apache Iceberg. Results are median of 5 runs.
End-to-end ingestion latency
Measured from Kafka message produce timestamp to first-readable Delta table commit on S3.
Test conditions
- Kafka 3.6 on 3-broker cluster
- Delta Lake table on S3 us-east-1
- Exactly-once semantic enabled
- 256-byte average message size
- 4 streaming compute nodes
- 30-second micro-batch window
~28% lower egress cost on 50TB datasets
Eliminating the copy-to-warehouse step reduces cross-AZ and egress traffic. On 50TB test dataset, measured savings were in the ~28% range vs a copy-to-warehouse approach running on equivalent infrastructure.
AWS us-east-1 egress + cross-AZ pricing, 50TB hot dataset, 30-day measurement period. Actual savings vary by cloud provider, region, and access pattern.
Run these benchmarks yourself.
Free tier includes 500 GB scan / month — enough to try the TPC-DS queries on a sample dataset.