https://hgpu.org/?p=9109
Scaling up scientific computations by using map-reduce-like control flow on NUMA architectures