Performance regressions

Hello,

I've been doing some benchmarking on recent version of PostgreSQL and
I'm seeing some regressions. The benchmark setup is as described in [1]
except it looks like I got lucky in the runs used for that article.

After many more runs, I get these NOPM averages (hopefully formatting
will survive):

Users: 50 100 250 500
12.4 485,914 707,535 739,808 589,856
13.0 485,501 697,505 837,446 385,225
14(2020-10-13) 521,902 759,609 941,212 611,647
14(2020-11-02) 478,640 684,138 930,959 513,707

The 14s are taken from the nightly builds on those dates, I can't find
any way to associate them with a specific git hash.

For 50 and 100 users, it seems there isn't all that much difference.
250 users keeps improving, but at 500 users the wheels seem to fall off
in 13, coming back a bit in 14.

I'm not sure exactly what to make of this but we thought it would be
important to raise the issue with the community. I'm trying to bisect
the 13 development cycle to see if anything stands out as the culprit,
but that will take some time.

I plan to keep doing 14 at the start of every month (I started late in
October).

[1]
https://www.enterprisedb.com/blog/postgresql-tproc-c-benchmarks-postgresql-12-vs-postgresql-13-performance

--
Vik Fearing