Skip to main content
. 2017 May 16;11:30. doi: 10.3389/fninf.2017.00030

Figure 2.

Figure 2

Map of memory consumption at the end of the network construction described in Figure 1. For parallelization with OpenMP (top) 48 threads are used and, correspondingly, for MPI (bottom) 48 processes (rightmost data points in Figure 1). The absolute contributions to the total memory consumption are distinguished by color from left to right. In the case of OpenMP these are: synapses (blue, 2 GiB), neurons (red, 72 MiB), overhead at start of program (yellow, 6.4 MiB), initialization of MPI with MPI_Init() (green, 9.4 MiB), and other NEST data structures (black, 7.9 MiB). MPI_Init() is not required for OpenMP parallelization and shown for comparison with MPI parallelization only. The enlargement (indicated by gray lines) shows the latter three contributions (total 23.7 MiB) multiplied by 48 (1138 MiB). In the case of the MPI simulation the data are sums over all processes, the last three components occupying a total of 1184 MiB. Memory consumption is measured by the resident set size (see Section 2.5).