SEVEN KEYS FOR PRACTICAL UNDERSTANDING AND USE OF CGNS Marc - PowerPoint PPT Presentation

SEVEN KEYS FOR PRACTICAL UNDERSTANDING AND USE OF CGNS Marc Poinot, SAFRAN Christopher Rumsey, NASA LaRC AIAA SciTech 2018

Motivation CGNS is well known in the CFD community, but: Common comments: CGNS is a library - often mismatched with its mid-level library CGNS cannot fit my own data structure - only understood as mesh storage on disk There are too many ways to describe the same feature in CGNS CGNS cannot handle parallel processing CGNS is only for archival data We have seen some initiatives to develop a new standard to: CGNS files are too big - define an implementation-independent data model The recommended implementation is HDF5 but only ADF is available There are few useful tools for CGNS manipulation - add Computer-Aided Design (CAD) related features I already have HDF5 and I do not need CGNS - implement in HDF5 CGNS inefficiently stores time data and statistics è This looks quite close to CGNS We propose to gather initiatives rather than creating entropy: let us refresh our comprehension… AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

CGNS entities and their acronyms Acronym Meaning Entity CG CGNS CFD General Notation System Identifier CG CGNS/SC Steering Committee Group of people CG CGNS/SIDS Standard Interface Data Structure Textual document CGNS Proposal for EXtension Procedure and textual document CGNS/CPEX CG CG CGNS/FMM File Mapping Manual Textual document CG CGNS/HDF5 HDF5 implementation Textual document CG CGNS/Py Python Python implementation Textual document CGNS/MLL CG Mid-Level Library Software library AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Seven keys 1 Abstract model > What is the data I want to define? > How do I define it? 2 Interoperability > How to go from CAD to mesh, to post-processing, visualisation and then back to CAD? > How to exchange data during CFD/CSM, CFD/CAA or any multiphysics workflows? 3 Functional coverage > Do I have enough functions to describe my data? > What is the remaining volume data I need to define/store? 4 High performance > Is there a High Performance Computing (HPC) efficient representation for my data model? > Who is going to support it? 5 Versatile implementation > Can I manipulate my data in a flexible way? > Do I have to re-invent all the functions I need for my data processing? 6 Extensibility > How can I describe out-of-CFD data, such as chemistry, high order, CSM features, new CFD trends? > Does the standard belong to NASA? 7 Open System > Can I connect the simulation elements together as black boxes? > Is there a mandatory implementation? AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Abstract model Types and Indexing conventions A model is what you want to specify SI SIDS S type definitions but not how you want to implement it Formal definition, supported using a grammar Informal definition, textual non-exhaustive description AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Interoperability The highest level of interoperability is the user's meaning The lowest is the implementation Families are the means to hold user’s information CFD/Flows • Topology entities, CAD reference • Same entity for different models • Entities with special features • Simulation specialized entities Simultaneous families allowed Hierarchy of families è Insure traceability through processes è Allows high level interoperability Motion/Deformation CAD/Topology CFD/CSM/FSI/Experimental AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

99% of the volume of the data already is defined Functional coverage The CGNS tree changes all along the simulation workflow > It covers our CFD needs and more > It doesn't cover everything, but it could eventually fulfill any needs: CPEX BaseIteration/ZoneIteration links CAD POST PRE CFD Family DimensionalUnits UserDefinedData Solution/ZoneSubRegion GridLocation/DataArray Base Zone/Grid Periodicity/Symmetry GoverningEquations BC AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

High performance HPC implementation > The HDF5 implementation specification is public > HDF5 w Write your own application w HDF Group support for portage/tuning w Specific node orientation w Very large data but limited number of types w Specific memory mapping (interlaced, non-contiguous, partial, out-of-core…) w Fast memory/disk swap w Smart memory mapping w Parallel features AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Versatile implementation Gluing/prototyping the easy way > Python/numpy w Your C/Fortran array IS your python array w Reuse of the very large python libraries w No-class paradigm, easy serialization/deserialization > Memory to memory exchange (same process, inter-process, inter-nodes, inter-hosts…) > Fast creation/modification of CGNS trees Change all BCs to family defined AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Extensibility user has a need check CGNS/SIDS • A support for user needs check CGNS/CPEX > CPEX submit informal email on w Propose your own extension cgnstalk@lists.nasa.gov w Do not break already used features create informal working group > Examples submit more formal proposal w NGON HPC redefinition w Arbitrary High order motivation/ rationale/ examples w Enhanced Chemical data model CGNS/SIDS update proposal w Arbitrary Reference Frame CGNS/FMM update proposal > PROS CGNS/MLL update proposal w Arguing often leads to better ideas review w Once adopted, everyone uses your data model / implementation > CONS CGNS/SC vote w Lengthy process CGNS/SIDS release w You have to explain your ideas CGNS/MLL release AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Open system Specification & connection only through public interfaces > High Level: CGNS/SIDS > Low Level: CGNS/FMM > Implementation: CGNS/HDF5, CGNS/Python > Any compliant implementation can 'connect' to others > Dedicated implementation is required for HPC CGNS/HDF5 CGNS/ADF CGNS/Python single-node mappings your own implementations CGNS/FMM CGNS/SIDS per-node data model CGNS/MLL pyCGNS mapping CHLone AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.

Answers to remarks CGNS is a library ý CGNS cannot fit my own data structure ý XML is the way to describe data ý Ex Extensi nsion n pr process ss is s too long ng for me þ Th There ar are e too man many ways to des escrib ibe e the e same ame fea eature e in in CGN GNS þ CGNS cannot handle parallel processing ý HPC cannot waste time on data model ý CGNS is only for archival data ý CG CGNS files files ar are e too big ig þ The recommended implementation is HDF5 but only ADF is available ý There ar Th are e few usefu ful l tools ls for CGN GNS man manip ipula latio ion þ I already have HDF5 and I do not need CGNS ý CG CGNS in ineffic ficien iently ly stores es time ime data a an and st statist stics þ CGNS is only for CFD CG ¨ AIAA-SciTech 2018 / Seven keys for practical understanding and use of CGNS Ce document et les informations qu’il contient sont la propriété de Safran. Ils ne doivent pas être copiés ni communiqués à un tiers sans l’autorisation préalable et écrite de Safran.