This repo is a testbed for benchmarking basic geospatial queries across a range of data stores.
It is written in Ruby, and organized as series of Rake tasks.
Perform a set of 100 bounding box queries that represent a moving window, as might be fetched from a front-end map client, over a swath of the United States extending from New York to Florida.
Repeat this N times for each data store.
The spatial dataset for this benchmark is the Museum Universe Data File, published in 2018 by the Institute of Museum and Library Services, a collection of ~30,000 museums and related organizations in the United States.
For N=10, i.e. 1000 bounding box queries, here are the results, based on wall-clock time:
| Data store | Version | Index | Elapsed (real) | Normalized | Throughput |
|---|---|---|---|---|---|
| Elasticsearch | 8.12 | n/a | 1.74 sec | 1.0 | 574 queries/sec |
| Postgres / PostGIS | 15.4 / 3.3 | GiST | 6.20 sec | 3.56 | 161 queries/sec |
| MongoDB | 7.0 | 2dsphere | 65.79 sec | 37.8 | 15.2 queries/sec |
Hardware note: This is on a 2023-vintage Mac laptop:
- MacBook Pro
- Apple M2 Pro, 12-core, 3.49GHz
- 32 GB RAM
You will need a working Docker installation.
This project includes a docker-compose configuration that will bring up services for:
- PostgreSQL with the PostGIS spatial extensions
- MongoDB
- Elasticsearch
- Clone this project and install its dependencies
git clone https://github.com/anandaroop/spatial-benchmarks.git
cd spatial-benchmarksasdf installbundle install- Obtain the MUDF csv datafile
bundle exec rake get_csv- Start the servers
docker-compose up- Create the spatial database and indexes, and load up the data
rake load- Run the benchmarks
rake benchmark