News for hackers. Stuff that matters.
Re:Dear Postgres by justinclift on Friday October 13, 2017 @03:43PM (#15468718)
Keep in mind that the "performance advantage" MySQL supposedly "had" over PostgreSQL mostly came down to a bad default setting in PG. Back in early PG versions (7.x from memory), the default shared buffers setting value was extremely low, in order to allow PG to "just work" even on very low memory systems. (8MB?)

The effect of not changing that from the default was PG then pegging the cpu at 100% (moving stuff into and out of the tiny buffer) with terrible performance.

When we raised that default to something reasonable for (at the time) modern systems... our performance magically increased substantially from the perspective of many non-DBAs.

Note - The exact details are in the PG email archives if anyone cares enough to look. :)

Re:Dear Postgres by rgbrenner on Friday October 13, 2017 @04:46PM (#15469121)
NO. Nothing I wrote above has anything to do with what you've written here. (It's so irrelevant, I have to ask: did you read my post at all?)

PG 7.x was 2000-2003. A full 5 years after MySQL was released.. and 7 years after mSQL.

The performance of mSQL versus pre-Postgres95 has nothing to do with a setting in PG 7.x released years later.

The performance advantage of mSQL was directly related to the stripped down DBMS that was developed to replace Postgres (AGAIN IN 1993) and the incomplete SQL implementation (AGAIN IN 1993)... both tradeoffs that Postgres did not make.

The issue you're referring to is a separate performance issue that did not occur until years later.. after MySQL already won the marketshare race.

Between the start of Postgres in 1986 and PG 7.x, there's about 14 years of history you're forgetting.

Re:Dear Postgres by anarazel on Friday October 13, 2017 @04:07PM (#15468867)
I think that's not quite right. There's a lot of things besides default settings that lead to not so great performance. Before spread checkpoints, HOT, short varlenas, readonly transactions not needing a txid, the free space map (remember max_fsm_pages), autovacuum, custom plans for prepared statements, asynchronous commits... it was a lot harder to use PG for higher throughput systems.
Re:Dear Postgres by justinclift on Friday October 13, 2017 @04:14PM (#15468900)
Those likely contributed, but the shared_buffers setting was the outright major culprit at the time.

Go look back through the -hackers mailing list, probably ~2001-2002. Or maybe just look at the timestamps for when the default changed, in order to get the right time range. :)

Re:Dear Postgres by anarazel on Friday October 13, 2017 @04:32PM (#15469034)
I used postgres before and after all of these changes, and I used mysql at that time too...

And before you accuse me of anti postgres bias: I work on postgres all day :)