Ben: On Fri, 8 Oct 2021 at 13:14, talk to ben <blo.talkto@xxxxxxxxx> wrote: > I am playing around with the gist fast build and comparing the result between v13 and v14. ... > When I compare the performance with the following script, I get a lot more data read into the buffers in v14 and a little slower query. ... > Is it expected ? (is the test dumb / too artificial ?) I'm not knowlegeable enough in gist or low level stuff, but I would suggest you run the same scripts in both, just in case the copy/create is influencing anything. I mean, once you have the scripts, do .. > \timing on > CREATE TABLE gist_fastbuild AS SELECT point(random(),random()) as pt FROM generate_series(1,10000000,1); > CREATE INDEX ON gist_fastbuild USING gist (pt); > VACUUM ANALYZE gist_fastbuild; > \di+ gist_fastbuild_pt_idx .. > COPY gist_fastbuild TO '/tmp/gist_fastbuild.copy'; On a scratch database, and then run the same script: > --pg13 > \timing on > CREATE TABLE gist_fastbuild(pt point); > COPY gist_fastbuild FROM '/tmp/gist_fastbuild.copy'; > CREATE INDEX ON gist_fastbuild USING gist (pt); > VACUUM ANALYZE gist_fastbuild; > \di+ gist_fastbuild_pt_idx > EXPLAIN (ANALYZE, BUFFERS) SELECT pt FROM gist_fastbuild WHERE pt <@ box(point(.5,.5), point(.75,.75)); a couple times or more on each server to rule artifacts out. ( testing speed in general, which such fast times, I myself would doubt comparisons of anything but exact runs ) It may give the same results, but then you know it does. Francisco Olarte.
