Re: GSoC proposal: Fast GiST index build

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Mon, Apr 4, 2011 at 7:16 AM, Alexander Korotkov <aekorotkov(at)gmail(dot)com> wrote:
>> Project name
>> Fast GiST index build

> Would/could/should this be implemented in a manner similar to the
> existing "GIN fast update" feature?

Fast build and fast update tend to be two different problems ...

> It's occurred to me to wonder whether even btree indexes would benefit
> from this type of optimization.

GIN fast update is a win when you can optimize the insertion of multiple
occurrences of the same key. There isn't really any corresponding
optimization possible in btree, AFAICS. (Heikki did some work awhile
back on btrees with multiple TIDs per key, for low-cardinality tables,
which might conceivably admit of a similar optimization. But I haven't
heard anything about that in a long time. It wasn't real clear to me
where the win over GIN would be for that.)

regards, tom lane