The 7th Annual Meeting on High Performance Computing and Infrastructure in Norway

NOTUR2008 - ABSTRACTS

Large-Scale Parallel Simulation of Neuronal Networks

Hans E. Plesser

Progress in understanding the brain depends critically on the simulation of models of neuronal networks to explore hypotheses about brain function. Demands on computational power for such simulations is growing strongly at present, as the focus in neuroscience research shifts to self-regulating and learning processes in the brain, which require not only more complex models, but also several orders of magnitude longer simulated times.

From a computer science perspective, large-scale neuronal network simulations pose interesting challenges: Network nodes communicate only through delayed, stereotypical pulses, allowing for embarrasingly parallel simulation interrupted by occasional complete exchange of compact messages bewtween parallel processes. The delivery of such messages to network nodes on each process, though, can exhaust memory bandwidth due to highly irregular access to very large adjacency lists.

In my talk, I will give an overview of current trends in neuronal network simulation and existing simulation software. I will give special attention to the NEST simulator co-developed by UMB (www.nest-initiative.org) and present benchmark results for several architectures.

< Back to programme