Computational & Technology Resources
an online resource for computational,
engineering & technology publications
Civil-Comp Proceedings
ISSN 1759-3433
CCP: 107
PROCEEDINGS OF THE FOURTH INTERNATIONAL CONFERENCE ON PARALLEL, DISTRIBUTED, GRID AND CLOUD COMPUTING FOR ENGINEERING
Edited by:
Paper 31

Recent and Upcoming Changes in Code_Saturne: Computational Fluid Dynamics HPC Tools Oriented Features

Y. Fournier1, J. Bonelle1, E. Le Coupanec1, A. Ribes2, B. Lorendeau2 and C. Moulinec3

1EDF, R&D/MFEE, Chatou, France
2EDF, R&D/SINETICS, Clamart, France
3Science and Technology Facilities Council, Daresbury, United Kingdom

Full Bibliographic Reference for this paper
Y. Fournier, J. Bonelle, E. Le Coupanec, A. Ribes, B. Lorendeau, C. Moulinec, "Recent and Upcoming Changes in Code_Saturne: Computational Fluid Dynamics HPC Tools Oriented Features", in , (Editors), "Proceedings of the Fourth International Conference on Parallel, Distributed, Grid and Cloud Computing for Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 31, 2015. doi:10.4203/ccp.107.31
Keywords: computational fluid dynamics, PRACE, Code_Saturne, meshes, unstructured, HPC, petascale.

Summary
In recent years, several advances have been brought to the HPC features of the Code_Saturne tool, through EDF developments and also through various PRACE projects, as Code_Saturne is part of the restricted Unified European Benchmark Application Suite of software. Several of the advances have been described in recent papers. In this work, we will focus on another set of new features relevant to HPC, dealing with IO and post-processing, and will also describe the expected impact of the addition of new compatible discrete operator methods to the code, which is currently mainly finite-volume based.

In a previous paper, we have described the type of operation required by the code to handle single-image IO. Here, we will describe the re-factoring in progress of these operations, and how a user-selectable and extensible handling of the underlying all-to-all operations allows better instrumentation, choice of algorithm, and future optimizations. Results using both MPI Alltoallv and Crystal Router will be compared.

We will also briefly present the new in-situ post-processing features, based on ParaView, which allow avoiding the need for writing huge amounts of data, and from having to transfer that data across systems.

Finally, we will describe the upcoming addition of compatible discrete operator discretization and their further impact on the code's linear solver resolution, notably from a matrix structure and numbering perspective. These operators and finite volume algorithms should coexist in the code for many years, so both will need to be managed, and future optimization efforts will need to account for the CDO operator, both regarding specific requirements, and representative performance use cases.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £45 +P&P)