Achieving Memory Scalability in the Gysela Code to Fit Exascale Constraints

Fabien Rozar (1, 2) , Guillaume Latu (1) , Jean Roman (2)

1 IRFM - Institut de Recherche sur la Fusion par confinement Magnétique
2 HiePACS - High-End Parallel Algorithms for Challenging Numerical Simulations

Gyrokinetic simulations lead to huge computational needs. Up to now, the semi-Lagrangian code Gysela performed large simulations using a few thousands cores (65k cores). But to understand more accurately the nature of the plasma turbulence, finer resolutions are wished which make Gysela a good candidate to exploit the computational power of future Exascale machines. Among the Exascale challenges , the less memory per core issue is one of the must critical. This paper deals with memory management in order to reduce the memory peak, and presents an approach to understand the memory behaviour of an application when dealing with very large meshes. This enables us to extrapolate the behaviour of Gysela for expected capabilities of Exas-cale machine.

Mots clés: Exascale; Memory scalability; Memory footprint reduction; Plasma Physics

Domaines: Calcul parallèle, distribué et partagé [cs.DC]

Fabien Rozar, Guillaume Latu, Jean Roman. Achieving Memory Scalability in the Gysela Code to Fit Exascale Constraints. PPAM 2017 - 10th International Conference on Parallel Processing and Applied Mathematics, R. Wyrzykowski et al., Sep 2013, Warsaw, Poland. pp.185-195, ⟨10.1007/978-3-642-55195-6_17⟩. ⟨hal-00935519⟩