Optimizing Vlasiator for European extreme-scale systems: porting and performance analysis on SiPearl Rhea HPC microprocessor

Abstract

This master's thesis aims to improve the efficiency of Vlasiator, the world's first simulation model that uses the Vlasov equation to model Earth's plasma environment in its global context. The research focuses on porting the application to Rhea, the high-performance / low-consumption microprocessor that SiPearl is developing, and that is being designed to be used for European extreme-scale systems. This work aims to compile Vlasiator to ARM, the base architecture behind Rhea, and identify the possible bottlenecks present in the application. The application's hotpots will be analyzed to identify potential performance improvements. The final goal is to compare the performance gains achieved using Scalable Vector Extension (SVE) and Vector Length Agnostic (VLA) implementations against traditional approaches. Future research directions may explore alternative approaches to overcome the identified limitations and enhance performance by exploiting other techniques.