On Sat, Dec 6, 2008 at 8:38 AM, Merlin Moncure <mmoncure@gmail.com> wrote:
> So what do we gather from this? Well, the feature works as
> advertised. I think that as long as your updates are not uniformly
> distributed across pages, vismap is a huge performance win for many
> workloads. I think the benefit will increase as the feature is
> tweaked in future versions. vacuum times are one of the things that
> make dealing with large tables difficult, and force us to use
> partitioning (which is, frankly, a hack).
>
> Why are new pages initialized dirty? Do inserts on pages set the dirty bit?
dumb question...there is no guarantee the transaction will be committed.
merlin