Distributed, Parallel, and Cluster Computing
Extension du prototypage rapide pour les pipelines d’imagerie radioastronomique : Simulation avancée multi-nœud HPC
Published on - 30e Colloque sur le traitement du signal et des images
SimSDP is a rapid prototyping tool designed to provide early estimates of the computation time and power consumption of SKAO’s future SDP processor pipeline. This paper extends a previous proof-of-concept by: (1) exploring more parallelism through dataflow modeling of spectral parallelism, (2) simulating pipeline behavior on multi-core, multi-node CPU architectures, (3) improving performance modeling and incorporating automation of the computation time estimation process. The results show that this tool facilitates the orientation of design decisions by exploring three radio-interferometric imaging algorithms on HPC systems, with an improvement in the reliability of the simulations, the uncertainty being reduced to around 10 %.