July 7, 2026
Benchmarks, but make it messy
PostgreSQL Benchmark: AWS RDS vs. Self-Hosted on Hetzner (2026)
Cheap cloud showdown turns into a comments-section food fight over a ‘rigged’ database race
TLDR: The test said Hostim beat Amazon’s small managed database on write speed, while Hetzner won on reads and Amazon looked pricey. Commenters, however, were far more interested in accusing the benchmark of being stacked, biased, or straight-up marketing dressed as research.
A benchmark meant to answer a simple money question — is Amazon’s managed database worth it compared with cheaper rivals? — ended up detonating a full-on comment war. The test claimed Hostim’s managed PostgreSQL service crushed Amazon RDS on write speed and beat a basic self-run setup on Hetzner too, while Hetzner won on reads. Translation for normal humans: one option looked best at saving changes fast, another looked best at fetching info fast, and Amazon looked expensive and oddly slow.
But the real fireworks were in the replies. Critics immediately yelled that the match-up felt like comparing a tuned sports car, a factory default sedan, and a rental with the parking brake half on. One commenter basically called the Hetzner comparison “useless,” because the self-run server was left on plain default settings while the managed services came pre-optimized. Another side-eyed the use of Amazon’s burstable budget machine, arguing it can slow down during longer tests, making the result feel stacked for drama.
Then came the scorched-earth reviews. One reader dismissed the whole thing as “self-promoting slop,” while another hit with the classic: “what could go wrong” when a company selling managed databases publishes a database comparison. And yet not everyone was booing — one commenter flexed that modern AI tools now make running your own highly reliable database far less scary, basically turning the thread into a surprise DIY rebellion. So yes, there were numbers — but the real headline was trust, bias, and whether this benchmark was science or marketing in a lab coat.
Key Points
- •The article benchmarked PostgreSQL 16 in July 2026 across Hostim managed Postgres, AWS RDS db.t4g.medium, and a self-hosted Hetzner CPX22 instance, all sized at 2 vCPU and 4 GB RAM.
- •In the write-heavy pgbench test, Hostim recorded 2,708 throughput at 1.48 ms latency, versus 1,303 at 3.07 ms for Hetzner and 1,080 at 3.71 ms for AWS RDS.
- •In the read-only pgbench test, Hetzner led with 20,068 at 0.40 ms, followed by Hostim at 14,333 and AWS RDS at 13,261.
- •The methodology used separate 4 vCPU load generators in the same region, a pgbench scale factor of 50, 300-second read and write runs, and a 60-second single-connection write test.
- •The article attributes much of the write-performance gap to network-attached block storage on the commit path and notes that the AWS db.t4g.medium instance is burstable, which can affect sustained CPU performance.