Simulation Neurotechnologies for Advancing Brain Research: Parallelizing Large Networks in NEURON.

Area of Science:

• Computational Neuroscience
• Neuroscience Technology

Background:

• Large-scale neuronal network simulations are crucial for understanding brain organization, driven by big data initiatives like Brain Research through Advancing Innovative Neurotechnologies (BRAIN).
• Developing effective simulation technologies is essential for handling the increasing complexity and scale of these models.

Purpose of the Study:

• To evaluate the performance of the NEURON simulator using Message Passing Interface (MPI) on High-Performance Computers (HPCs) for moderately large neuronal networks.
• To benchmark simulation efficiency across different network sizes and node configurations.

Main Methods:

• Utilized the NEURON simulator with MPI on the Neuroscience Gateway's HPC resources.
• Simulated neuronal networks ranging from 500 to 100,000 cells across 1 to 256 nodes.
• Compared simulation performance for Izhikevich integrate-and-fire (I&F) neurons, single-compartment Hodgkin-Huxley (HH) cells, and hybrid networks.

Main Results:

• Simulation run time demonstrated a near-linear increase with network size.
• Simulation run time showed a near-linear decrease with an increasing number of nodes.
• Networks composed of I&F neurons exhibited slightly faster run times compared to HH networks, with minimal differences due to single-compartment cell models.

Conclusions:

• The NEURON simulator with MPI provides a scalable solution for large neuronal network simulations on commonly available HPCs.
• Simulation performance scales predictably with network size and parallelization, supporting big data neuroscience research.
• The choice between I&F and HH single-compartment neuron models has a minor impact on simulation speed in this context.