Figure 2

Impact of grid map reuse, OpenMP multithreading, and I/O on mpAD4 execution speed. (A) A 32,000 copy indinavir library was docked with mpAD4 on a 2,048(8,192) node(core) BG/P system in VN node mode (MPI + OpenMP with 1 OpenMP thread/node, 4× virtual nodes) with either grid map reloading or reuse and SMP mode (MPI + OpenMP with 4 OpenMP threads/node) with grid map reloading or reuse. The system was configured with 64 I/O nodes (1 I/O node per 128 cores). Runtimes were normalized to the fastest case (SMP, gm = reuse). (B) An 8,000 copy indinavir library was docked on a 1,024(4,096) node(core) BG/P system in VN mode (gm = reload or gm = reuse) and SMP mode (gm = reuse) configured with either 8 I/O nodes (I/O poor) or 64 I/O nodes (I/O replete). Runtimes were normalized to the fastest case (SMP, gm = reuse, I/O = 64).