GoKawiilPostgreSQL blogger revisits random_page_cost defaults after POSETTE talk
A PostgreSQL performance researcher shared follow-up findings on the database's random_page_cost setting, which governs how the query planner estimates the cost of random versus sequential disk reads. New benchmark data from older rotational SATA drives showed an estimated cost around 125, far higher than the long-standing default of 4.0, while SSD-based tests land closer to that default only by coincidence.
GoKawiil's interpretation of the reporting above, not reported fact.
The findings suggest PostgreSQL's decades-old default value was never actually derived from raw hardware measurements, raising questions about whether current tuning advice is built on a flawed premise. Users who report worse performance after raising random_page_cost may be inadvertently compensating for gaps in the planner's cost model, such as its failure to account for caching, rather than fixing an inaccurate hardware cost estimate.
- random_page_cost for old rotational drives measured near 125, far above PostgreSQL's default of 4.0
- SSD-based estimates land closer to 4.0, but only coincidentally, not because the default was designed for them
- The parameter may be masking flaws in the planner's cost model rather than reflecting true disk I/O costs
Samsung 990 PRO NVMe SSD — If this deep dive into random_page_cost convinced you that spinning SATA disks are a relic, the Samsung 990 PRO is the modern counterpoint: an M.2 NVMe drive whose random reads land far closer to sequential than any rotational disk ever could. It's a great fit for a home Postgres test rig where you want to benchmark planner cost settings against real flash storage.
See Samsung 990 PRO NVMe SSD on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: vondra.me, 2026-09-04
Published there as: “Some more thoughts on random_page_cost”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.