Re: For index bloat: VACUUM ANALYZE vs REINDEX/CLUSTER

Richard Broersma Jr wrote:
> --- Gregory Williamson <Gregory.Williamson@digitalglobe.com> wrote:
>
>
>> A very low fill factor means that pages are
>> "sparse" and so inserts and updates are less likely to trigger massive b-tree rebalancings.
>>
>
> I take it that "massive b-tree rebalancings" could cause a problem with the performance of disk
> writing though-put from UPDATEs and INSERTs?
>
> Regards,
> Richard Broersma Jr.
>
Precisely -- even if it can keep everything in RAM it can occupy quite a
few cycles to rebalance a large b-tree. And eventually those changes do
need to get written to disk so the next checkpoint (I think) will also
have more work.
G