> In other words, cache misses are taken into account to the extent that the model above is reflective of your workload and hw environment. Fair enough. Though

> In other words, cache misses are taken into account to the extent that the model above is reflective of your workload and hw environment. Fair enough. Though I think they have in mind cache misses in the sense of page faults here? - i.e. "it's not in memory - have to read from disk". (i.e. cache as in Postgres internal cache (+ OS filesystem cache, actually/IIRC). The variable name `random_page_cost` seems to suggest this (load a memory page into memory - from disk). Whereas I had in mind cache as in CPU cache[1] - where sort operation happens which does lots of cache misses (between CPU cache and RAM) - and needs to read from rest of RAM (i.e. random memory access is not uniformly random). It's a nice rabbit hole.. [1]: https://en.wikipedia.org/wiki/CPU_cache https://en.wikipedia.org/wiki/CPU_cache