Benchmarking

Instancing benchmark

Sciview can display geometry in a number of ways. The normal approach is to add a node to the scene for each object you want to see. This has the advantage that you can easily manipulate and delete individual nodes from the scene. However, this does not scale well for use cases where you need to populate your scene with hundreds or thousands of the same or similar looking objects such as spheres, cones or cylinders.

Instancing allows you to render the same object from memory to multiple positions on the screen, greatly reducing render times at the expense of flexibility. You can still transform and translate each instance individually, and even give them custom properties to pass to the shader, but you lose the ability to interact with a single instance in the scene.

To test the speed differences between the conventional node publishing approach and instanced rendering, Sciview comes with a benchmark that populates the scene with a number of icospheres. You can run this benchmark as a gradle task from the command line, from a Sciview window (menu Demo > Advanced) or by running the task in Intellij IDEA from:

Gradle Tasks > demos.advanced > runInstancingBenchmark

To call the benchmark from the command line use:

Bash (Linux, MacOS, git bash)

./gradlew runInstancingBenchmark -Dsciview.benchmark.numParticles=20 -Dsciview.benchmark.type=instanceSequential

Windows Powershell

gradlew.bat runInstancingBenchmark "-Dsciview.benchmark.numParticles=20" "-Dsciview.benchmark.type=instanceSequential"

There are four benchmark types available:

Type	Description	Time for 1000 spheres
publishAll	Publishes a node added event after every single sphere creation	20 seconds
publishOnce	Suppresses the node added event and publishes all changes cumulatively at the end	1 second
instanceSequential	Uses instancing to add spheres to the scene. The loop is performed sequentially.	0.1 second
instanceParallel	Uses instancing to add spheres to the scene. The loop is performed in parallel by turning it into a parallel stream	< 0.1 seconds

The execution time is measured in milliseconds and logged to the console.