On 17 January 2013 15:14, Heikki Linnakangas <hlinnakangas@vmware.com> wrote:
> On 17.01.2013 16:53, Robert Haas wrote:
>>
>> On Thu, Jan 17, 2013 at 3:49 AM, Pavan Deolasee
>> <pavan.deolasee@gmail.com> wrote:
>>>
>>> May be you've already addressed that concern with the proven
>>> performance numbers, but I'm not sure.
>>
>>
>> It would be nice to hear what Heikki's reasons were for adding
>> PD_ALL_VISIBLE in the first place.
>
>
> The idea was to avoid clearing the bit in the VM page on every update, when
> the bit is known to not be set, ie. when the PD_ALL_VISIBLE flag is not set.
> I assumed the traffic and contention on the VM page would be a killer
> otherwise. I don't remember if I ever actually tested that though. Maybe I
> was worrying about nothing and hitting the VM page on every update is ok.
Presumably we remember the state of the VM so we can skip the re-visit
after every write?
-- Simon Riggs http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services