On Wed, 2006-05-03 at 10:17 -0400, Tom Lane wrote:
> Simon Riggs <simon@2ndquadrant.com> writes:
> You are optimizing the wrong thing here. If we choose not to mark an
> entry dead then we will pay for that omission on every future scan of
> the same entry. I don't think that outweighs the (doubtless rare)
> situation where we expend an extra page fetch to reload the page.
Sounds a familiar conversation, which I shouldn't have raised here.
This depends upon whether the pages being accessed are in cache or not,
and whether we have sufficient I/O to pay the cost of a write. Reads
don't always go to disk, writes always do. I see that its difficult to
tell which is which, but that doesn't mean there aren't different cases.
--
Simon Riggs
EnterpriseDB http://www.enterprisedb.com