spatial_index_benchmark
spatial_index_benchmark copied to clipboard
mloskot
Simple non-academic performance comparison of available open source implementations of R-tree spatial index using linear, quadratic and R* balancing algorithms as well as bulk loading.
Results and graphs could be automatically generated and published by Travis CI using the [Travis GitHub Deployer](http://www.boost.org/community/groups.html#users) utility.
A Benchmark for Multidimensional Index Structures by Bernhard Seeger http://www.mathematik.uni-marburg.de/~seeger/rrstar/index.html
Currently, queries are only based on spatial predicates defining relationship between stored values and some geometry. Queries for NN predicate defining relationship between stored values and some point need to...
Currently, we test random rectangles. It would be interesting to see results for, let's say, 10M random points.
Currently, all times include time spent on construction of indexeable values which next are indexed (`boost::geometry::box` for BGI and `SpatialIndex::Region` for LSI). In order to get even more accurate benchmarks...
It would be interesting to compare in-memory use of the [R-Tree Extension](http://gist.cs.berkeley.edu/libgist-2.0/libgist_user_manual.html#usingrtext) implemented in [libgist2](http://gist.cs.berkeley.edu/libgist-2.0/) from the [GiST](http://gist.cs.berkeley.edu) project.
Add benchmark for the [Efficient and Lightweight In-Memory Implementation of R*-Tree](http://www.ics.uci.edu/~salsubai/rstartree.html) implemented by Sattam Alsubaiee. We will call this library RST.
The "Force Reinsert" (see the linked paper) can be configured for Boost.Geometry `rtree`, so various values should be profiled. At least, comparison of with and without force reinsert is needed....