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

This slide illustrates the distribution of connected components when utilizing a static and deterministic approach. The distribution is based solely on the ID assigned to different MPI processes. As shown, the distribution is not ideal; several MPI processes, particularly those associated with higher values, have few or no connected components.

Despite this, the overall repartition is adequate for our algorithm's scalability, particularly from a memory perspective. This demonstrates that the algorithm can effectively handle both weak and strong scaling scenarios.