I chose project "Indexing prolonged geometrical objects (i.e. boxes, circles, polygons, not points) with SP-GiST by mapping to 4d-space".
After meeting with Alexander Korotkov, I wrote some plan. What are postgresql developers thinking about it?
According to the presentation SP-GIST 3 times faster than GiST in some cases. But GIST supports geometrical data types: box, circle, polygon with operators: && &> &< &<| >> << <<| <@ @> @ |&> |>> ~ ~=
Popular spatial extension PostGIS doesn't include SP-GIST, but has a lot of geometrical features.
Using of K-D-tree and Quadtree in building index for geometrical data types can increase speed of search in some cases.
The main idea is representing 2-D geometrical objects in their bounding box. Set of 2-D boxes is 4-D space. New _ops will work with points from 4-D space, for example kd_box_ops, quad_circle_ops and will support all geometrical operators.
After conversion object to their bounding box algo has set of tuples (x1, y1, x2, y2). Our goal is separate this space the most equally.
If we talk about K-D-tree, on first step K-D-tree algorithm will split space in 2 parts by the first coordinate, in next step by the second coordinate etc., after 4-th coordinate we repeat this procedure. At the end we have index at geometrical objects and use traversal tree for every search operator.
Postgresql has already has realization ideas of MBR in gist/gistproc.c. So I will transfer this realization to other type of tree.
Of cource, I assume that SP-GIST can be not the best decision of this problem. So after testing this clear methods, I will try to find more effective way. Maybe with using combination of different spatial tree structures.