Nathan Boley <npboley@gmail.com> writes:
>> The accesses to an index are far more likely to be clustered than the
>> accesses to the underlying table, because the index is organized in a
>> way that's application-meaningful and the table not so much.
> So, to clarify, are you saying that if query were actually requesting
> rows uniformly random, then there would be no reason to suspect that
> index accesses would have hotspots? It seems like the index structure
> ( ie, the top node in b-trees ) could also get in the way.
The upper nodes would tend to stay in cache, yes, but we already assume
that in the index access cost model, in a kind of indirect way: the
model only considers leaf-page accesses in the first place ;-)
regards, tom lane