Supabase launches DBArena, open benchmarks comparing hosted Postgres providers

DBArena is a benchmark site for hosted Postgres, and Supabase sponsors it. The first release compares Supabase Postgres with default heap storage and with OrioleDB (beta), Amazon RDS for PostgreSQL, and Google Cloud SQL for PostgreSQL Enterprise edition. Each product runs at up to seven compute tiers, matched by vCPU count, on Postgres 17 with each provider's default settings and direct TLS connections without a pooler.

Each tier is tested in two scenarios: one where the data fits in memory with a buffer cache hit ratio of at least 99%, and one where the data is four times the available RAM, forcing disk reads. Results report transactions per minute, transactions per minute per dollar of monthly on-demand list price, and p95 latency. Each test has a 30-minute warmup and a 60-minute measurement, and the published figure is the median of three successful runs at peak concurrency.

The workload is derived from TPC-C with think time disabled, so it measures maximum throughput. The project states that it does not meet all TPC-C requirements and its results are not comparable to official TPC-C results. Each published result has a result.json file in the dbarenact repository on GitHub with its configuration, price, software versions and exact command, and tests so far run only in AWS us-east-1 and GCP us-east1.

Teams choosing a Postgres backend for agent workloads can now compare providers at matched vCPU counts and normalized cost, and rerun any test point with the published open source tools.

Because the workload uses no think time, runs in one region per cloud, and uses list prices without discounts, the numbers show peak capacity rather than typical agent traffic, so results should be checked against your own access pattern.