Figure 4.

Speed up versus number of MPI processes for a post-processing ensemble run of a test system (192 replicas, 7622 atoms, 30000 frames) consisting of stripping solvent molecules, re-imaging the remaining atoms, and optionally writing output trajectories. All tests were run on 192 nodes. Black line with circles: old setup where each MPI process accessed each member of the ensemble during setup; output trajectories written. Blue line with squares: new setup where only the master MPI process for each ensemble member performs setup; output trajectories written. Orange line with diamonds: new setup, do not write output trajectories. Calculations run on NCSA Blue Waters XE6 nodes.