Neurokernel: An Open Source Platform for Emulating the Fruit Fly Brain.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neuroinformatics

Background:

• The fruit fly Drosophila melanogaster brain is a complex system with a well-defined modular structure.
• Collaborative modeling efforts require platforms that can integrate diverse, independently developed computational modules.
• Efficient execution and testing of large-scale brain models necessitate high-performance computing resources like Graphics Processing Units (GPUs).

Purpose of the Study:

• To develop an open software platform, Neurokernel, for the collaborative modeling and simulation of the fruit fly brain.
• To facilitate the integration of independently developed models of brain modules.
• To leverage multi-GPU architectures for efficient execution and performance benchmarking.

Main Methods:

• Developed Neurokernel, an open software platform with a programming model based on the modular organization of the fruit fly brain.
• Defined mandatory communication interfaces for inter-module data exchange, abstracting internal model designs.
• Integrated independently developed models of the retina and lamina neuropils.
• Utilized direct GPU-to-GPU data transfers for communication between modules.

Main Results:

• Demonstrated successful integration of independently developed models of the fruit fly's visual system (retina and lamina).
• Showcased the neuroinformation processing capabilities of the integrated models.
• Benchmarked Neurokernel's communication performance, demonstrating scalability with an increasing number of modules and interface ports on multi-GPU systems.

Conclusions:

• Neurokernel provides a powerful framework for collaborative, large-scale brain modeling.
• The platform effectively integrates diverse models by standardizing communication interfaces.
• Neurokernel efficiently utilizes multi-GPU architectures, showing scalable communication performance crucial for complex brain simulations.