PostgresBench: A Reproducible Benchmark for Postgres Services
4 hours ago
- #database benchmarking
- #managed Postgres
- #performance evaluation
- Performance is a core design goal, applied to both ClickHouse (analytical) and managed Postgres (transactional), forming a unified data stack.
- PostgresBench is introduced as a public, reproducible benchmark for managed Postgres services, similar to ClickBench for OLAP databases, using pgbench with TPC-B-like workloads.
- Benchmark parameters include 256 clients, 16 threads, 10-minute runs, and two scale factors (100 GB and 500 GB) to test performance under different data sizes and storage pressures.
- Results are reported for five services (Postgres managed by ClickHouse, AWS Aurora, AWS RDS, Neon, Crunchy Bridge) across small and large instance configurations, showing metrics like TPS, average latency, P95, and P99.
- The methodology emphasizes fairness with consistent infrastructure, default Postgres configurations, and transparent documentation, though pricing comparisons are excluded as ClickHouse's managed Postgres was unreleased at testing time.