Technical Note: Parallel implementation of the EGSnrc Monte Carlo simulation of ionizing radiation transport using...

Area of Science:

• Computational physics
• Monte Carlo simulations
• Scientific computing

Background:

• The EGSnrc Monte Carlo system is a widely used tool for simulating radiation transport.
• Existing parallel processing methods for EGSnrc rely on batch queuing systems (BQS), which can be cumbersome and less efficient.
• There is a need for more flexible and performant parallelization strategies for complex scientific simulations.

Purpose of the Study:

• To implement and evaluate a new parallel processing option for the EGSnrc Monte Carlo system using the OpenMP API.
• To provide a portable and efficient alternative to the existing BQS-based parallelization method.
• To assess the performance and accuracy of the OpenMP implementation (OMP_EGS) against the serial and BQS methods.

Main Methods:

• The OMP_EGS solution was developed by integrating OpenMP features into the original EGSnrc source code using macros.
• Validation involved comparing OMP_EGS results with the serial implementation to ensure exact agreement.
• Performance was evaluated by measuring parallel speedup and efficiency against the BQS method.

Main Results:

• The OMP_EGS implementation allows for easy activation or deactivation of OpenMP features, enabling direct recovery of the serial implementation.
• Validation tests confirmed that OMP_EGS reproduces the exact same results as the serial EGSnrc implementation.
• Performance and scalability tests demonstrated that OMP_EGS offers superior runtime and parallel efficiency compared to the EGSnrc BQS parallel implementation.

Conclusions:

• OMP_EGS presents a more compact and portable parallel processing solution for the EGSnrc system compared to BQS.
• The OpenMP implementation is compatible with various compilers and operating systems, enhancing its usability across different environments.
• OMP_EGS provides better performance and parallel efficiency without introducing artifacts into simulated distributions.