Laurenz Albe <laurenz.albe@cybertec.at> writes:
> On Mon, 2025-10-06 at 01:29 -0400, Tom Lane wrote:
>> But if what
>> we're trying to model is net resource demands, with an eye to
>> minimizing the total system load not execution time of any one query,
>> maybe we can continue to work with something close to what we've
>> traditionally done.
> Did anybody propose that?
I just did ;-). If we don't adopt a mindset along that line,
then AIO is going to require some *radical* changes in the
planner's I/O cost models.
> I was under the impression that PostgreSQL's cost model tries to
> estimate and optimize execution time, not resource consumption.
Yup, that's our traditional view of it. But I wonder how we
will make such estimates in a parallel-I/O world, especially
if we don't try to account for concurrent query activity.
(Which is a place I don't want to go, because it would render
planning results utterly irreproducible.)
> But perhaps I misunderstood, or perhaps I am just too conservative.
I'm normally pretty conservative also about changing planner
behavior. But in this context I think we need to be wary of
thinking too small. The fact that people keep coming out with
different ideas of what random_page_cost needs to be suggests
that there's something fundamentally wrong with the concept.
regards, tom lane
