mirror of https://github.com/postgres/postgres
657499dbab
picksplit algorithm for Rtree. Please apply patch below to CVS Oleg Bartunov |
||
---|---|---|
.. | ||
bench | ||
data | ||
expected | ||
sql | ||
Makefile | ||
README.rtree_gist | ||
rtree_gist.c | ||
rtree_gist.sql.in |
README.rtree_gist
This is R-Tree implementation using GiST. Code (for PG95) are taken from http://s2k-ftp.cs.berkeley.edu:8000/gist/pggist/ and changed according to new version of GiST (7.1 and above) All work was done by Teodor Sigaev (teodor@stack.net) and Oleg Bartunov (oleg@sai.msu.su). See http://www.sai.msu.su/~megera/postgres/gist for additional information. CHANGES: Oct 10 MSD 2001 1. Implemented new linear algorithm for picksplit ref. ( 'New Linear Node Splitting Algorithm for R-tree', C.H.Ang and T.C.Tan ) Tue May 29 17:04:16 MSD 2001 1. Small fixes in polygon code Thanks to Dave Blasby <dblasby@refractions.net> Mon May 28 19:42:14 MSD 2001 1. Full implementation of R-tree using GiST - gist_box_ops,gist_poly_ops 2. gist_poly_ops is lossy 3. NULLs support 4. works with multi-key GiST NOTICE: This version will works only with postgresql version 7.1 and above because of changes in interface of function calling. INSTALLATION: gmake gmake install -- load functions psql <database> < rtree_gist.sql REGRESSION TEST: gmake installcheck EXAMPLE USAGE: create table boxtmp (b box); -- create index create index bix on boxtmp using gist (b gist_box_ops); -- query select * from boxtmp where b && '(1000,1000,0,0)'::box; BENCHMARKS: subdirectory bench contains benchmark suite. Prerequisities: perl, DBI, DBD:Pg, Time::HiRes cd ./bench 1. createdb TEST 2. psql TEST < ../box.sql 3. ./create_test.pl | psql TEST -- change $NUM - number of rows in test dataset 4. ./bench.pl - perl script to benchmark queries. Run script without arguments to see available options. a)test without GiST index, using built-in R-Tree ./bench.pl -d TEST b)test R-Tree using GiST index ./bench.pl -d TEST -g RESULTS: 1. One interesting thing is that insertion time for built-in R-Tree is about 8 times more than ones for GiST implementation of R-Tree !!! 2. Postmaster requires much more memory for built-in R-Tree 3. Search time depends on dataset. In our case we got: +------------+-----------+--------------+ |Number boxes|R-tree, sec|R-tree using | | | | GiST, sec | +------------+-----------+--------------+ | 10| 0.002| 0.002| +------------+-----------+--------------+ | 100| 0.002| 0.002| +------------+-----------+--------------+ | 1000| 0.002| 0.002| +------------+-----------+--------------+ | 10000| 0.015| 0.025| +------------+-----------+--------------+ | 20000| 0.029| 0.048| +------------+-----------+--------------+ | 40000| 0.055| 0.092| +------------+-----------+--------------+ | 80000| 0.113| 0.178| +------------+-----------+--------------+ | 160000| 0.338| 0.337| +------------+-----------+--------------+ | 320000| 0.674| 0.673| +------------+-----------+--------------+