Reproducible performance results
HarborSQL vs Databricks SQL benchmarks
HarborSQL benchmarks measure client-observed query time and modeled compute cost across ClickBench, point lookups, SSB, and Delta type compatibility. The raw results, runners, and cost assumptions are public so you can inspect every claim.
Latest ClickBench result
HarborSQL v0.1.9 on an AWS c8i.2xlarge completed the 43-query optimized ClickBench suite with a 91.351-second best total. The published Databricks 2X-Small baselines completed in 165.7 seconds with Cloud Fetch and 170.3 seconds without it.
On this benchmark, HarborSQL used 44.9% to 46.4% less client time on the best-total measure. Its modeled server compute cost was $0.0113, compared with $0.1657 to $0.1703 for the Databricks baselines. These numbers describe this dataset and topology, not every production query.
| System | Compute | Best total | Median total | Modeled best cost |
|---|---|---|---|---|
| HarborSQL v0.1.9 | AWS c8i.2xlarge | 91.351s | 92.852s | $0.0113 |
| Databricks v2026.10 | 2X-Small SQL Warehouse | 170.3s | 192.1s | $0.1703 |
| Databricks v2026.10 with Cloud Fetch | 2X-Small SQL Warehouse | 165.7s | 179.5s | $0.1657 |
Methodology and hardware
- Run the benchmark client in the same cloud region as the data.
- Measure wall-clock time observed by the client.
- Execute each ClickBench query three times.
- Compare an AWS c8i.2xlarge HarborSQL host with a managed Databricks 2X-Small SQL Warehouse.
- Use an AWS t3.small benchmark client for the published comparison.
Published datasets
- ClickBench: 43 analytical queries over an optimized S3-backed hits table.
- Point lookup: concurrent lookups against the ClickBench table.
- SSB SF10: a star-schema decision-support workload.
- Delta types: result compatibility across common Delta and Arrow types.
How to read these results
Best total sums the fastest observed run for each query. Median and average totals show how repeat runs behaved. Cost figures use server or warehouse compute only and exclude the benchmark client, storage, network transfer, monitoring, support, taxes, idle time, and the labor required to operate HarborSQL.
HarborSQL is a single-engine, read-only runtime. Databricks SQL Warehouse includes managed scaling and a broader SQL and result delivery surface. Benchmark your own query mix, concurrency, data layout, and cloud region before making a production decision.