https://hgpu.org/?p=12188
Revisiting Edge and Node Parallelism for Dynamic GPU Graph Analytics