Enabling parallel computing in CRASH

TL;DR

The paper introduces pCRASH2, a parallelized version of the CRASH2 radiative transfer code, optimized for supercomputers, enabling efficient simulation of radiation from many sources with validated performance improvements.

Contribution

It presents a new parallelization strategy for CRASH2 that ensures linear speed-up and efficient handling of multiple radiation sources in cosmological simulations.

Findings

Achieves linear speed-up with increasing cores for equal or more sources than cores

Effectively propagates radiation from many sources in parallel

Validated through extensive test cases confirming performance and accuracy

Abstract

We present the new parallel version (pCRASH2) of the cosmological radiative transfer code CRASH2 for distributed memory supercomputing facilities. The code is based on a static domain decomposition strategy inspired by geometric dilution of photons in the optical thin case that ensures a favourable performance speed-up with increasing number of computational cores. Linear speed-up is ensured as long as the number of radiation sources is equal to the number of computational cores or larger. The propagation of rays is segmented and rays are only propagated through one sub-domain per time step to guarantee an optimal balance between communication and computation. We have extensively checked pCRASH2 with a standardised set of test cases to validate the parallelisation scheme. The parallel version of CRASH2 can easily handle the propagation of radiation from a large number of sources and is ready for the extension of the ionisation network to species other than hydrogen and helium.