On-the-fly clustering for exascale molecular dynamics simulations.

For weak scaling, our algorithm demonstrates some scalability, although it is slightly less effective compared to other algorithms. This is partly due to our focus on developing a robust and portable algorithm for future GPU integration. However, our algorithm still exhibits lower performance than the initial attempt from the PHGT.

In terms of strong scaling results, we observe promising scalability despite the presence of a large connected component. This central component increases the number of iterations required as the number of MPI processes rises. Nevertheless, we maintain decent scaling properties.

We achieved scalability with up to 4,400 cores. Notably, the memory footprint of this algorithm remains nearly constant across different MPI processes, regardless of the number of processes utilized.