Action Exploratoire INRIA ExODE

Coordinateur: Jonathan Rouzaud-Cornabas (INRIA Beagle, Liris)

Participants: Samuel Bernard (INRIA Dracula, Institut Camille Jordan), Thierry Gautier (Avalon)

Date: 2019-2022

En Français:
En biologie, la grande majorité des systèmes peut être modélisée sous la forme d’équations différentielles ordinaires (ODE). Modéliser plus finement des objets biologiques mène à augmenter le nombre d’équations. Simuler des systèmes toujours plus grands mène également à augmenter le nombre d’équations. Par conséquent, nous observons une explosion de la taille des systèmes d’ODE à résoudre. Un verrou majeur est la limitation des logiciels de résolutions numériques d’ODE (solveur ODE) à quelques milliers d’équations à cause de temps de calcul prohibitif. L’AEx ExODE s’attaque à ce verrou via 1) l’introduction de nouvelles méthodes numériques qui tireront parti de la précision mixte qui mélange plusieurs précisions de nombre flottant au sein d’un schéma de calcul, 2) l’adaptation de ces nouvelles méthodes pour des machines de calcul de prochaines générations qui sont fortement hiérarchiques et hétérogénes et composées d’un grand nombre de CPUs et GPUs. Depuis un an, une nouvelle approche du Deep Learning se propose de remplacer les Recurrent Neural Network (RNN) par des systèmes d’ODE. Les méthodes numériques et parallèles d’ExODE seront évalué et adapté dans ce cadre afin de permettre l’amélioration de la performance et de l’exactitude de ces nouvelles approches.

En Anglais:
In biology, the vast majority of systems can be modeled as ordinary differential equations (ODEs). Modeling more finely biological objects leads to increase the number of equations. Simulating ever larger systems also leads to increasing the number of equations. Therefore, we observe a large increase in the size of the ODE systems to be solved. A major lock is the limitation of ODE numerical resolution software (ODE solver) to a few thousand equations due to prohibitive calculation time. The AEx ExODE tackles this lock via 1) the introduction of new numerical methods that will take advantage of the mixed precision that mixes several floating number precisions within numerical methods, 2) the adaptation of these new methods for next generation highly hierarchical and heterogeneous computers composed of a large number of CPUs and GPUs. For the past year, a new approach to Deep Learning has been proposed to replace the Recurrent Neural Network (RNN) with ODE systems. The numerical and parallel methods of ExODE will be evaluated and adapted in this framework in order to improve the performance and accuracy of these new approaches.

Avalon is contributing to LLVM OpenMP/runtime

Philippe Virouleau, funding thanks to EoCoE-II project, has proposed patches to LLVM OpenMP/runtime in order to provide better control and performances of OpenMP task execution. First accepted patch was pushed in LLVM master branch (https://reviews.llvm.org/D63196). It solves an side effect due to the task throttling heuristic that serializes task execution. It may  cripple the application performance in some specific task graph scenarios, like the ones detailed in section 4.2 from this paper published at IWOMP 2018 (the full text can be found here). In such cases not having the full task graph prevent some opportunities for cache reuse between successive tasks.

Mid term goals are to transfer some important and innovative features mostly already available in libKOMP https://gitlab.inria.fr/openmp/libkomp.

 

 

EoCoE-II

Summary

Europe is undergoing a major transition in its energy generation and supply infrastructure. The urgent need to halt carbon dioxide emissions and prevent dangerous global temperature rises has received renewed impetus following the unprecedented international commitment to enforcing the 2016 Paris Agreement on climate change. Rapid adoption of solar and wind power generation by several EU countries has demonstrated that renewable energy can competitively supply significant fractions of local energy needs in favourable conditions. These and other factors have combined to create a set of irresistible environmental, economic and health incentives to phase out power generation by fossil fuels in favour of decarbonised, distributed energy sources. While the potential of renewables can no longer be questioned, ensuring reliability in the absence of constant conventionally powered baseload capacity is still a major challenge.

The EoCoE-II project will build on its unique, established role at the crossroads of HPC and renewable energy to accelerate the adoption of production, storage and distribution of clean electricity. How will we achieve this? In its proof-of-principle phase, the EoCoE consortium developed a comprehensive, structured support pathway for enhancing the HPC capability of energy-oriented numerical models, from simple entry-level parallelism to fully-fledged exascale readiness. At the top end of this scale, promising applications from each energy domain have been selected to form the basis of 5 new Energy Science Challenges in the present successor project EoCoE-II that will be supported by 4 Technical Challenges

Partners
CEA, FZJ, ENEA, BSC, CNRS, INRIA, CERFACS, MPG, FRAUNHOFER, FAU, CNR, UNITN, PSNC, ULB, UBAH, CIEMAT, IFPEN, DDN, RWTH, UNITOV

Project Information
EoCoE-II is a H2020 RIA european project, call H2020-INFRAEDI-2018-1.

Duration: 3 years, Jan 1st 2019, Dec 31st 2021.

Avalon Members: T. Gautier, C. Perez

Online Resources

URL: https://www.eocoe.eu/