On Tue, Aug 17, 2021 at 11:24 AM Bruce Momjian <bruce@momjian.us> wrote:
> OK, I am trying to think of something simple we could test to see the
> benefit, with few downsides. I assume the case you are considering is
> that you have a 10 8k-page table, and one page is 80% full and the
> others are 81% full, and if several backends start adding rows at the
> same time, they will all choose the 80%-full page.
That's one problem (there is so many). They start with that same page,
and then fight each other over other pages again and again. This
emergent behavior is a consequence of the heuristic that has the FSM
look for the heap page with the least space that still satisfies a
given request. I mentioned this problem earlier.
> For example:
>
> 1. page with most freespace is 95% free
> 2. pages 2,4,6,8,10 have between 86%-95% free
> 3. five pages
> 4. proc id 14293 % 5 = 3 so use the third page from #2, page 6
Right now my main concern is giving backends/XIDs their own space, in
bulk. Mostly for concurrent bulk inserts, just so we can say that
we're getting the easy cases right -- that is a good starting point
for my prototype to target, I think. But I am also thinking about
stuff like this, as one way to address contention.
> This should spread out page usage to be more even, but still favor pages
> with more freespace. Yes, this is simplistic, but it would seem to have
> few downsides and I would be interested to see how much it helps.
I thought about having whole free lists that were owned by XIDs, as
well as shared free lists that are determined by hashing MyProcPid --
or something like that. So I agree that it might very well make
sense.
--
Peter Geoghegan
