Re: I/O on select count(*)

Greg Smith escribió:
> On Thu, 15 May 2008, Pavan Deolasee wrote:
>
>> I had suggested in the past that whenever we set hint bits for a tuple,
>> we should check all other tuples in the page and set their hint bits
>> too to avoid multiple writes of the same page. I guess the idea got
>> rejected because of lack of benchmarks to prove the benefit.
>
> From glancing at http://www.postgresql.org/docs/faqs.TODO.html I got the
> impression the idea was to have the background writer get involved to
> help with this particular situation.

The problem is that the bgwriter does not understand about the content
of the pages it is writing -- they're opaque pages for all it knows. So
it cannot touch the hint bits.

I agree with Pavan that it's likely that setting hint bits in batches
instead of just for the tuple being examined is a benefit. However,
it's perhaps not so good to be doing it in a foreground process, because
you're imposing extra cost to the client queries which we want to be as
fast as possible. Perhaps the thing to do is have a "database-local
bgwriter" which would scan pages and do this kind of change ...
a different kind of process to be launched by autovacuum perhaps.

If we had the bitmask in a separate map fork, this could be cheap.

--
Alvaro Herrera http://www.CommandPrompt.com/
The PostgreSQL Company - Command Prompt, Inc.