Re: WIP: Avoid creation of the free space map for small tables

Hi John Naylor,
On Tue, Jan 8, 2019 at 2:27 AM John Naylor <jcnaylor@gmail.com> wrote:
> I've attached two patches for testing. Each one applies on top of the
> current patch.

Thanks for the patch, I did a quick test for both of the patches same
tests as in [1], now for fillfactors 20, 70, 100 (Note for
HEAP_FSM_CREATION_THRESHOLD = 4 highest tid inserted was 20 fillfactor
is (3,43),  for 70 fillfactor is (3, 157) and for 100 fillfactor is
(3, 225), so exactly 4 pages are used)

Machine : cthulhu, same as before [2] and server settings is default.
Test: COPY command as in [1], for 500 tables.

Fill factor 20
                      execution time in ms            %increase in
execution time
Base                             119.238
v11-all-pages                121.974                 2.2945705228
v11-Every-other-page   114.455                 -4.0113051209
v11-last-page                113.573                 -4.7510021973

Fill factor 70
                     execution time in ms           %increase in execution time
Base                              209.991
v11-all-pages                 211.076                0.5166888105
v11-Every-other-page    206.476              -1.6738812616
v11-last-page                 203.591              -3.0477496655

Fill factor 100
                   execution time in ms            %increase in execution time
Base                              269.691
v11-all-pages                 270.078                 0.1434975583
v11-Every-other-page    262.691                -2.5955630703
v11-last-page                 260.293                -3.4847288193

Observations
1. Execution time of both base and v11-all-pages patch has improved
than my earlier results [2]. But still v11-all-pages is slightly
behind base.
2. v11-Every-other-page and v11-last-page patches improve the
performance from base.
3. IMHO v11-Every-other-page would be ideal to consider it improves
the performance and also to an extent avoid expansion if space is
already available.

[1] https://www.postgresql.org/message-id/CAJVSVGX%3D2Q52fwijD9cjeq1UdiYGXns2_9WAPFf%3DE8cwbFCDvQ%40mail.gmail.com
[2] https://www.postgresql.org/message-id/CAD__Ouj%3Dat4hy2wYidK90v92qSRLjU%2BQe4y-PwfjLLeGkhc6ZA%40mail.gmail.com

-- 
Thanks and Regards
Mithun Chicklore Yogendra
EnterpriseDB: http://www.enterprisedb.com