Routing Brain Traffic through the Bottlenecks of General Purpose Computers: Challenges for Spiking Neural Network Simulation Code
Performance Analysis and Optimization
TimeTuesday, June 18th3:45pm - 4:05pm
DescriptionSimulation of spiking neuronal networks on general purpose computers is a third pillar supporting the investigation of the dynamics and function of biological neural networks, next to experimental and theoretical approaches. In recent years, ever more scalable simulation code has been developed that works efficiently for a broad variety of models and on various platforms, from laptops to supercomputers. The challenges for neuronal simulators on general purpose computers arise from the sparse but broad connectivity in neuronal network models and from the unpredictable pulse signaling of neurons, called spiking. In distributed simulations of such networks, this requires frequent communication of spike data from varying sources to varying targets, which is a principle complication for the efficient exchange of data between processes. Moreover, routing of the spike data received by a process to the local targets entails irregular memory access, which constitutes a major performance bottleneck. My talk will address such challenges for spiking network simulators and present recent developments in simulation technology that aim at meeting these challenges.