The purpose of this project is to demonstrate the feasibility and performance of a realtime geofencing advertising backend.
Polygons with 6 sides the size of a few blocks are recorded in a rectangle representing approximately America.
Each polygon is linked to a document stored by Kuzzle in Redis.
A request to the API allows to know if given GPS coordinates are contained in one of the polygons. If this is the case, then the saved document corresponding to the polygon is returned by Kuzzle.
Requests to the API are authenticated.
This benchmark is realised with a Kuzzle cluster on AWS m5.large instances.
Server specifications: 2 vCPU, 8GB RAM
- Number of 6 vertices polygons:
300 000 - Overlaping polygons:
true - Accuracy:
<1m - Zone:
USA - Kuzzle authentication:
yes - Document storage:
Redis - Protocol:
Websocket - Node.js:
8.11.0
The test consists in repeating the same request 2000 times with a point matching 1 polygon.
The benchmark is realized with Gatling and a websocket scenario.
Gatling server is a c5.2xlarge instance.
Server specifications: 8 vCPU, 16 GB RAM
| Kuzzle nodes | Concurrent users | Requests / second | Requests / second / user | Requests / second / node | Latency (ms) | Full benchmark |
|---|---|---|---|---|---|---|
| 2 | 20 | 4700 | 235 | 2350 | 3 | 80 users in 40 seconds |
| 2 | 120 | 6000 | 50 | 3000 | 17 | 120 users in 12 seconds |
| 2 | 200 | 7000 | 35 | 3500 | 31 | 200 users in 20 seconds |
| 3 | 120 | 8500 | 71 | 2833.33 | 12 | 120 users in 12 seconds |
| 3 | 190 | 9000 | 47 | 3000 | 16 | 200 users in 20 seconds |
| 3 | 200 | 10000 | 50 | 3333.33 | 21 | 200 users in 10 seconds |
| 4 | 120 | 10000 | 83 | 2500 | 8 | 120 users in 12 seconds |
| 4 | 180 | 11000 | 61 | 2750 | 11 | 200 users in 20 seconds |
| 4 | 200 | 11000 | 55 | 2750 | 19 | 200 users in 10 seconds |
| 4 | 400 | 15000 | 38 | 3750 | 26 | 400 users in 20 seconds |
Number of request per second during 400 concurrent user benchmark
Cluster and Gatling server load during 400 concurrent user benchmark
This benchmark is realised with a standalone Kuzzle stack on a Scaleway C2L server.
Server specifications: 8 dedicated CPU cores, 32GB RAM, SSD, 600Mb/s network
- Number of 6 vertices polygons:
300 000 - Overlaping polygons:
true - Accuracy:
<1m - Zone:
USA - Kuzzle authentication:
yes - Document storage:
Redis - Protocol:
Websocket - Node.js:
8.11.0
The test consists in repeating the same request 2000 times with a point matching 1 polygon.
The benchmark is realized with Gatling and a websocket scenario .
Server specifications: 4 dedicated CPU cores, 8GB RAM, SSD, 300Mb/s network
| concurrent connections | avg latency (ms) | avg request/s | max requests/s |
|---|---|---|---|
| 1 | 3 | 222 | 388 |
| 2 | 3 | 400 | 641 |
| 3 | 4 | 546 | 757 |
| 4 | 6 | 602 | 853 |
| 5 | 6 | 667 | 980 |
| 10 | 10 | 910 | 1231 |
| 20 | 16 | 1144 | 1455 |
Same test but progressively load 120 users with 200 requests each.
The full benchmark report is available here
This benchmark is realised with a standalone Kuzzle stack on a Scaleway C2L server.
Server specifications: 8 dedicated CPU cores, 32GB RAM, SSD, 600Mb/s network
- Number of 6 vertices polygons:
300 000 - Overlaping polygons:
true - Accuracy:
<1m - Zone:
USA - Kuzzle authentication:
yes - Document storage:
Redis - Protocol:
HTTP - Node.js:
8.11.0
The test consists in repeating the same request 2000 times with a point matching 1 polygon.
The benchmark is realised with bombardier on a Scaleway C2S server.
Server specifications: 4 dedicated CPU cores, 8GB RAM, SSD, 300Mb/s network
| concurrent connections | avg latency (ms) | request/s |
|---|---|---|
| 1 | 2.74 | 363 |
| 2 | 4.51 | 442 |
| 3 | 5.57 | 538 |
| 4 | 6.96 | 575 |
| 5 | 9.17 | 546 |
| 10 | 15.27 | 654 |
| 20 | 43.01 | 467 |
Randomly generate polygons and register them in geofencing filters.
Parameters:
count: number of polygons to registerbounding_box: area where to generate polygons
Use http://geojson.io to have bounding box coordinates.
- create a rectangle
- go to "Meta" menu
- click "add bounding box"
Bounding boxes:
- USA bounding box
[ -127.96875, 26.745610382199022, -66.4453125, 51.6180165487737]
Use the script actions/geofence-register.js to register 10 000 polygons in the USA.
Test if GPS coordinates are included in one of the registered polygons and return the corresponding documents.
Parameters:
lat: latitudelng: longitude
Example:
curl -H "Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJfaWQiOiJhc2NoZW4iLCJpYXQiOjE1NDMxMDUzMDQsImV4cCI6MTU0MzEwODkwNH0.WnUCDCwPXRUA1JE_4e7kbkIShQiM0MtW0admTWpKI1g" "http://localhost:7512/_plugin/geofencing-advertising/geofence/test?lat=-86.99962414458622&lng=31.431421096655942Get the polygons list in GeoJSON format.
Execute the script action/get-geojson.js to create a polygons.json file containing the polygons.
Use https://www.gmapgis.com/ to load the generated file in a world map.
This script send a request to register polygons filters.
Usage: node actions/geofence-register.js <host> <filter count> <bounding box>
The availables bounding boxes are: usa
Example: register 100 000 polygons filters in USA
node actions/geofence-register.js localhost 100000 usa
This script return random points that match at least one polygon.
It also print curl, bombardier and custom node benchmark command for each point.
Usage: node actions/geofence-find-match.js <host> <bounding box>
The availables bounding boxes are: usa
Example:
node actions/geofence-find-match.js localhost usa
This script get the registered polygons in GeoJSON format.
The generated file can be viewed online on https://www.gmapgis.com/, just drag'n'drop the file.
Usage: node actions/get-geojson.js <host> <filename>
Example:
node actions/get-geojson.js localhost ./polygons.json