Performance of hybrid programming models for multiscale cardiac simulations: preparing for petascale computation

Area of Science:

• Computational biology
• High-performance computing (HPC)
• Multiscale modeling

Background:

• Advancements in high-performance computing (HPC) are crucial for developing sophisticated multiscale and multiphysics models in medical research.
• Computationally efficient multiscale models necessitate advanced hybrid programming paradigms.

Purpose of the Study:

• To evaluate the performance of hybrid programming models for realistic physiological multiscale heart simulations.
• To compare Message Passing Interface (MPI) with hybrid MPI and multithreading approaches (OpenMP, Pthreads).

Main Methods:

• Implementation of a realistic physiological multiscale heart model.
• Performance comparison of MPI-only versus hybrid MPI/multithreading programming models.
• Utilizing OpenMP and Pthreads in conjunction with MPI.

Main Results:

• Hybrid programming models demonstrated superior performance compared to MPI-only implementations.
• The combination of OpenMP and MPI offered an effective balance between performance gains and code complexity.
• Hybrid models efficiently utilized all processor cores for both computation and communication.

Conclusions:

• Hybrid programming models, particularly OpenMP with MPI, are well-suited for accelerating multiscale physiological simulations on HPC systems.
• These models pave the way for achieving faster-than-real-time simulations, significantly advancing medical research and treatment development.
• Future HPC systems with increased core counts will further enhance the capabilities of these simulation techniques.