Tom:
Yes, they are ints. To (somewhat) check your guess on the role of the hash
aggregation speed, I just ran oltp test from sysbench
(http://sysbench.sourceforge.net/docs/#database_mode) on a table with 1mln
of records. That test uses pretty simple queries (that do not use
aggregation) and 8.3 showed about the same performance as 8.2 (strictly
speaking 8.3 was about 1-2% slower, but not 10-15% like on my query).
I'm curious if that new hash aggregation algorithm was put in 8.3 with the
performance increase as a goal or it was simply a required change to support
some other new feature of 8.3? Right now the switch from 8.2 to 8.3 doesn't
seems a favorable step for the type of application that we have...
-- vlad
----- Original Message -----
From: "Tom Lane" <tgl@xxxxxxxxxxxxx>
To: "Vlad" <marchenko@xxxxxxxxx>
Cc: "PG-General" <pgsql-general@xxxxxxxxxxxxxx>
Sent: Monday, January 28, 2008 9:13 PM
Subject: Re: 8.3RC2 vs 8.2.6 testing results
Vlad <marchenko@xxxxxxxxx> writes:
2. We ran several tests and found 8.3 generally 10% slower than 8.2.6.
The particular case you are showing here seems to be all about the speed
of hash aggregation --- at least the time differential is mostly in the
HashAggregate step. What is the data type of a_id? I speculate that
you're noticing the slightly slower/more complicated hash function that
8.3 uses for integers. On a case where the data was well distributed
you'd not see any countervailing efficiency gain from those extra
cycles.
regards, tom lane
---------------------------(end of broadcast)---------------------------
TIP 5: don't forget to increase your free space map settings
