Heikki Linnakangas wrote:
> Tom Lane wrote:
>> "Heikki Linnakangas" <heikki@enterprisedb.com> writes:
>>> I also wonder what the performance impact of extending BufferTag is.
>>
>> That's a fair objection, and obviously something we'd need to check.
>> But I don't recall seeing hash_any so high on any profile that I think
>> it'd be a big problem.
>
> I do remember seeing hash_any in some oprofile runs. But that's fairly
> easy to test: we don't need to actually implement any of the stuff,
> other than add a field to BufferTag, and run pgbench.
I tried that. hash_any wasn't significant on pgbench, but I was able to
construct a test case where it is. It goes like this:
BEGIN;
CREATE TEMPORARY TABLE foo (id int4);
INSERT INTO foo SELECT a FROM generate_series(1, 10000) a;
INSERT INTO foo SELECT * FROM foo;
INSERT INTO foo SELECT * FROM foo;
INSERT INTO foo SELECT * FROM foo;
... (repeat multiple times)
oprofile says that hash_any consumes ~7 % of CPU time on that test on my
laptop.
More precisely, on CVS HEAD it takes between 7.1-7.2%. After extending
BufferTag with one uint32, it takes 7.4-7.5%. So the effect is
measurable if you try hard enough, but not anything to get worried about.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com