Mathias Kunter <mathiaskunter(at)gmail(dot)com> writes:
> the documentation of the random_page_cost configuration parameter says:
>> Although the system will let you set random_page_cost to less than
>> seq_page_cost, it is not physically sensible to do so.
> However, I don't think this statement is true. Consider the situation
> where the randomly fetched pages are mostly (or even entirely) cached in
> RAM, but where the sequentially fetched pages must be mostly read from
> disk. An example for such a scenario is a database system which uses
> RAM-cached indices.
I think fooling with effective_cache_size is a better way to model
that situation.
regards, tom lane