Re: visibility map - what do i miss? - Mailing list pgsql-hackers

From Guillaume Smet
Subject Re: visibility map - what do i miss?
Date
Msg-id 1d4e0c10812060501j74e7a3c5h5f911715c7a4668d@mail.gmail.com
Whole thread Raw
In response to visibility map - what do i miss?  (hubert depesz lubaczewski <depesz@depesz.com>)
Responses Re: visibility map - what do i miss?  (Heikki Linnakangas <heikki.linnakangas@enterprisedb.com>)
List pgsql-hackers
On Sat, Dec 6, 2008 at 12:53 PM, hubert depesz lubaczewski
<depesz@depesz.com> wrote:
> First run - without visibility maps, timing of vacuums:
> Time: 267844.822 ms
> Time: 138854.592 ms
> Time: 305467.950 ms
> Time: 487133.179 ms
>
> Second run - on head:
>
> Time: 252218.609 ms
> Time: 234388.763 ms
> Time: 334016.413 ms
> Time: 575698.750 ms
>
> Now - as I understand the change - visilibity maps should make second run much faster?

If I understand correctly what Heikki explained, not if you run VACUUM
only once (and you confirmed me on IRC you run it only once). It's the
VACUUM which sets the PD_ALL_VISIBLE flag on the pages so the first
VACUUM should be slower with HEAD than with 8.3 as it has far more
work to do.

The second VACUUM should then be faster.

-- 
Guillaume


pgsql-hackers by date:

Previous
From: hubert depesz lubaczewski
Date:
Subject: visibility map - what do i miss?
Next
From: Heikki Linnakangas
Date:
Subject: Re: visibility map - what do i miss?