Thread: rtree: improve performance, tuple killing

This patch makes some improvements to the rtree index implementation:

(1) Keep a pin on the scan's current buffer and mark buffer. This avoids
the need to do a ReadBuffer() for each tuple produced by the scan.

(2) Convert a ReleaseBuffer() ; ReadBuffer() pair into
ReleaseAndReadBuffer(). Surely not a huge win, but it saves a lock
acquire/release...

(3) Remove a bunch of duplicated code in rtget.c; make rtnext() handle
both the "initial result" and "subsequent result" cases.

(4) Add support for index tuple killing

(5) Remove rtscancache(): it is dead code, for the same reason that
gistscancache() is dead code (an index scan ought not be invoked with
NoMovementScanDirection).

The end result is about a 10% improvement in index scan performance,
according to contrib/rtree_gist/bench.

These changes (with the exception of #2) are analogous to changes I've
already made for GiST (it's clear that GiST was started as a fork of
rtree). I'm not hugely interested in further improvements to rtree; I
just did this stuff because it is low-hanging fruit and I've already
made the same changes for GiST.

-Neil

Neil Conway <neilc@samurai.com> writes:
> This patch makes some improvements to the rtree index implementation:

Please hold this for the 8.1 cycle.  I think we are past the time for
non-bug-fix changes in the 8.0 release.

            regards, tom lane

On Tue, 2004-11-23 at 23:40 -0500, Tom Lane wrote:
> Please hold this for the 8.1 cycle.  I think we are past the time for
> non-bug-fix changes in the 8.0 release.

Sorry, I should have mentioned that explicitly: I have (and had) no
intention of applying this for 8.0.

-Neil

Barring any objections, I intend to apply this patch tomorrow. The
patch, as well as the original -patches email, are included below.

-Neil

On Wed, 2004-11-24 at 11:15 +1100, Neil Conway wrote:
> This patch makes some improvements to the rtree index implementation:
>
> (1) Keep a pin on the scan's current buffer and mark buffer. This avoids
> the need to do a ReadBuffer() for each tuple produced by the scan.
>
> (2) Convert a ReleaseBuffer() ; ReadBuffer() pair into
> ReleaseAndReadBuffer(). Surely not a huge win, but it saves a lock
> acquire/release...
>
> (3) Remove a bunch of duplicated code in rtget.c; make rtnext() handle
> both the "initial result" and "subsequent result" cases.
>
> (4) Add support for index tuple killing
>
> (5) Remove rtscancache(): it is dead code, for the same reason that
> gistscancache() is dead code (an index scan ought not be invoked with
> NoMovementScanDirection).
>
> The end result is about a 10% improvement in index scan performance,
> according to contrib/rtree_gist/bench.
>
> These changes (with the exception of #2) are analogous to changes I've
> already made for GiST (it's clear that GiST was started as a fork of
> rtree). I'm not hugely interested in further improvements to rtree; I
> just did this stuff because it is low-hanging fruit and I've already
> made the same changes for GiST.

On Tue, 2005-01-18 at 16:43 +1100, Neil Conway wrote:
> Barring any objections, I intend to apply this patch tomorrow.

Applied.

-Neil