Re: [Patch] Optimize dropping of relation buffers using dlist

On Wed, Dec 30, 2020 at 11:28 AM Tang, Haiying
<tanghy(dot)fnst(at)cn(dot)fujitsu(dot)com> wrote:
>
> Hi Amit,
>
> In last mail(https://www.postgresql.org/message-id/66851e198f6b41eda59e6257182564b6%40G08CNEXMBPEKD05.g08.fujitsu.local),
> I've sent you the performance test results(run only 1 time) on single table. Here is my the retested results(average by 15 times) which I think is more accurate.
>
> In terms of 20G and 100G, the optimization on 100G is linear, but 20G is nonlinear(also include test results on shared buffers of 50G/60G), so it's a little difficult to decide the threshold from the two for me.
> If just consider 100G, I think NBuffers/32 is the optimized max relation size. But I don't know how to judge for 20G. If you have any suggestion, kindly let me know.
>

Considering these results NBuffers/64 seems a good threshold as beyond
that there is no big advantage. BTW, it is not clear why the advantage
for single table is not as big as multiple tables with the Truncate
command. Can you share your exact test steps for any one of the tests?
Also, did you change autovacumm = off for these tests, if not then the
results might not be reliable because before you run the test via
Vacuum command autovacuum would have done that work?

--
With Regards,
Amit Kapila.