Comparison of computational load of a simple and complex electrophysiological cell models in large anatomical data-sets on the Blue Gene/L supercomputer

Abstract

Despite increasing computer power, long simulation times and memory resources remain limiting factors for use of large and detailed multi-scale cardiac models. We propose a modeling framework based on orthogonal recursive bisection and the MPI standard on the IBM Blue Gene/L supercomputer. We compute the cellular FitzHugh-Nagumo and ten Tuscher et al. cell model and monodomain equation based on a two ventricle model in 0.2 mm resolution to investigate the performance. The first 1000 time steps are calculated for comparison of all model weights on 512, 1024, 2048, 4096 and 8192 computational nodes. A 1 s (105 time steps) simulation was then carried out on 2048 nodes for comparison with other cardiac models. The speedup gained is linear up to 4096 nodes. Small differences in performance are noticeable between cell models used. The 1 s simulation is carried out in less than 50 minutes wall clock time. We could show that the simulation with long simulation times and detailed cardiac models can now be carried out within hours.