Scaling Results

For a simple supernova test calculation, we examine both the scaling and performance tradeoff of spatial versus wavelength parallelization. Figure 4 presents the results of our timing tests for one iteration of a Type Ic supernova model atmosphere, with a model temperature $\ifmmode{T_{\rm model}}\else{\hbox{$T_{\rm model}$} }\fi= 12,000$ K (the observed luminosity is given by $L=4\pi R^2 \ifmmode{T_{\rm model}}\else{\hbox{$T_{\rm model}$} }\fi^4$), characteristic velocity v₀=10000 $\,$km$\,$s^-1, 4666 NLTE levels, 163812 NLTE lines, 211680 LTE lines (for simplicity, all line profile were assumed to be Gaussian), non-homogeneous abundances, and 260630 wavelength points. This is a typical test for production calculations and we have designed this test to have the highest potential for synchronization, I/O waiting, and swapping to reduce performance to simulate a worst case scenario for the parallel performance. It is however, characteristic of the level of detail needed to accurately model supernovae. This calculation has also been designed to barely fit into the memory of a single node. The behavior of the speedup is very similar to the results obtained for test case using a model of a nova explosion [9]. The ``saturation point'' at which the wavelength pipeline fills and no further speedup can be obtained if more wavelength clusters are used for the machines used here, occurs at about 5 to 8 clusters. More clusters will not lead to larger speedups, as expected. Larger speedups can be obtained by using more worker nodes per cluster, which also drastically reduces the amount of memory required on each node.