Benchmarks of force-directed layouts

Benchmarks of force-directed layouts

Typically force-directed layouts contain multiple iterations and are therefore parallelizable. We would like to provide WebWorker / WASM and GPU approaches to obtain performance gains. Here's what we have done for now:

@antv/layout-gpu uses WebGL / WebGPU to implement GPGPU which has the best performance.
@antv/layout-wasm uses WebWorker multi-threaded shared memory to run WASM, with Rust rayon based parallelism behind it.

Choose a dataset:

Choose a layout:

Iterations:

Min movement:

Distance threshold mode: It means the condition to judge with min movement:

'mean' means the layout stops while the nodes' average movement is smaller than min movement
'max' / 'min' means the layout stops while the nodes' maximum/minimum movement is smaller than min movement

Choose test cases:

graphology

@antv/layout

@antv/layout-wasm singlethread

@antv/layout-wasm multithreads

@antv/layout-gpu

Scaling:

graphology

@antv/layout

@antv/layout-wasm singlethread

@antv/layout-wasm multithreads

@antv/layout-gpu