Figure 3

Scaling of MPI alone versus MPI with OpenMP multithreading. Virtual compound libraries were docked in VN and SMP modes on BG/P systems of 512(2,048), 1,024(4,096), 2,048(8,192) and 4,096(16,384) nodes(cores). (A) A 34,841 compound ZINC library was docked on variable BG/P systems sizes. Ideal speedup was calculated from the fastest 512 node result (SMP), and the relative speedup SMP and VN mode results were plotted. (B) A 9,000 compound subset ZINC library was docked on variable BG/P systems sizes, and ideal speedup was calculated from the fastest 512 node result (SMP). (C) A 4,000 copy indinavir library was docked on variable BG/P systems sizes, and an ideal speedup line was calculated from the fastest 512 node result (SMP). (D) Relative execution times for the 4,000 copy indinavir library were calculated for SMP and VN modes at each system size.