Grid'5000 user report for Sebastien Coudert

Jump to: navigation, search

User information

Sebastien Coudert (users, user, account-manager, lille, ml-users user)
More user information in the user management interface.

Experiments

  • PIV analysis software (C++) (Programming) [achieved]
    Description: PIV (Particle Image Velocimetry) analysis is based on image processing to measure velocity of a flow. The 2 components of the velocity are measured in a plane. The algorithm can be explicitely computed in parallel. The compressed input data flow is about 30 time bigger than the output result data flow.
    Analysis parallelisation possibility:
    • correlation windows (thread)
    • fields (process)


    language: C++
    documentation: doxygen
    libraries:
    • the CImgLibrary
    • FFTw(Fastest Fourier Transform in the West)
    • NetCDF(network Common Data Form)

    Results: Compile and run on standard system image.


  • PIV analysis (shell) (Application) [achieved]
    Description: Run PIV analysis software on grid.
    PIV (Particle Image Velocimetry) analysis is based on image processing to measure velocity of a flow.
    Parallelisation: shell script.
    Full analysis is splitted in different jobs, which consists in groups of single pair of images to process. Results are stored in separate files by each CPU. These files are concatenate after processing to a single one.
    Results: run PIVlml on grid: 256 stereoscopic image pairs processed.
    Both different compressions of the input data flow and number of CPU were used. Everything is running on main NFS disk.
    The processing seems not to be bounded by disk, but by number of CPU. Also, we switch to a bigger cluster.
    using idpot.grenoble
    nodes min. img.
    16 4 zlib
    8 10 raw
    22 3 raw

    nodes min. img.
    14 (16) 9 zlib
    7 (8) 19 zlib
    17 (22) 5 zlib
    15 (20) 6 zlib



  • SPIV analysis software (C++) (Programming) [in progress]
    Description: SPIV (Stereoscopic Particle Image Velocimetry) analysis is based on image processing to measure velocity of a flow. The 3 components of the velocity are measured in a plane.
    Analysis parallelisation possibility:
    • correlation windows (thread)
    • stereoscopic (thread)
    • fields (process)


    language: C++
    documentation: doxygen
    libraries:
    • the CImgLibrary
    • FFTw(Fastest Fourier Transform in the West)
    • NetCDF(network Common Data Form)
    • MPI (Message Passing Interface)

    Results:


  • SPIV analysis (shell) (Application) [achieved]
    Description: SPIV (Stereoscopic Particle Image Velocimetry) analysis is based on image processing to measure velocity of a flow.
    Parallelisation: shell script.
    Full analysis is splitted in different jobs, which consists in groups of 16 stereoscopic pair of images to process. Results are stored in separate files by each CPU. These files can be concatenate after processing to a single one.
    Results: For all experiences:
    - using lille cluster
    - size of about 50 GBytes input data flow (i.e. 9600 stereoscopic image pairs processed)
    - computations are working well using local directory to store results (i.e. /scratch/) and then transfer it to main NFS disk (i.e. using rsync)

    Test01
    - input data flow is on main NFS disk.
    - running on 60 nodes in about 2h of computation.
    - disk load is quite high, but it's working efficiently (i.e. CPU is almost at 100% all the time). But, main problem is that it may cause problem for other grid users.
    Disk tranfers about 15 MBytes/s out.

    Test02
    - input data flow is on main NFS disk.
    - running on 10 nodes in about 12h of computation.
    - disk load is less and it's working efficiently (i.e. CPU is almost at 100% all the time).
    Disk tranfers less than 10 MBytes/s out.

    For these previous cluster experiment, the processing seems to be bounded by disk access using the lille cluster.

    Test03
    - input data flow is on external file server (no more load of the main NFS disk).
    Network is about 10 MBytes/s in between external file server/nodes.
    Disk tranfers about 30 MBytes/s in.
    Network is about 30 MBytes/s in between nodes.
    Use FUSE/sshfs in between nodes in needed.
    DrawBacks: But files are some time removed.


  • SPIV analysis (MPI) (Application) [planned]
    Description: SPIV (Stereoscopic Particle Image Velocimetry) analysis software.
    Parallelisation: MPI.
    Full analysis is splitted in different jobs using MPI. Results are stored in a single file.
    Results:


  • SPIV analysis (shell) - Summary (Application) [in progress]
    Description: CALINS WallTurb/ Monash/acquisition-may Grid SPIV Rate Node CPU du (raw) grid nb field time data (Mo/s) SPIV (3C/s) total per node total per node total per node total per CPU total per CPU 32 64 320Go 20 Go 661x187 1936 121/2 96h 1h30 60 4 44k 700
    Results: CALINS WallTurb/ Monash/acquisition-may Grid SPIV Rate Node CPU du (raw) grid nb field time data (Mo/s) SPIV (3C/s) total per node total per node total per node total per CPU total per CPU 32 64 320Go 20 Go 661x187 1936 121/2 96h 1h30 60 4 44k 700
  • WallTurb FP05/Y2 - SPIV analysis (shell) (Application) [achieved]
    Description: WallTurb FP05/Y2
    TST10/XY
    TST10/YZ
    TST16/XY
    TST16/YZ

    Results: to do
    WallTurb/FP05/Y2
    Run Grid SPIV Rate
    Node CPU du (zlib)grid nb field time data (Mo/s) SPIV (3C/s)
    totalper node totalper node totalper node totalper CPU totalper CPU
    TST10/XY ?? ??? 48 Go 1? Go ???x??? 600x16 16x? ??h ?h?? ?? ? ??k ???
    TST10/YZ ?? ??? 117 Go 1? Go ???x??? 600x16 16x? ??h ?h?? ?? ? ??k ???
    TST16/XY ?? ??? 8 Go ? Go ???x??? 15?? 5?x2? ??h ?h?? ?? ? ??k ???
    TST16/YZ ?? ??? 9 Go ? Go ???x??? 15?? 5?x2? ??h ?h?? ?? ? ??k ???

  • WallTurb FP10/Y2 - SPIV analysis (shell) (Application) [achieved]
    Description: WallTurb FP10/Y2
    TST10/XY
    TST10/YZ
    TST16/XY
    TST16/YZ

    Results: to do
    WallTurb/FP10/Y2
    Run Grid SPIV Rate
    Node CPU du (zlib)grid nb field time data (Mo/s) SPIV (3C/s)
    totalper node totalper node totalper node totalper CPU totalper CPU
    TST10/XY ?? ??? ??Go 1? Go ???x??? 600x16 16x? ??h ?h?? ?? ? ??k ???
    TST10/YZ ?? ??? ??Go 1? Go ???x??? 600x16 16x? ??h ?h?? ?? ? ??k ???
    TST16/XY ?? ??? 5.5 Go 5.5 Go ???x??? 15?? 5?x2? ??h ?h?? ?? ? ??k ???
    TST16/YZ ?? ??? 5.5 Go 5.5 Go ???x??? 15?? 5?x2? ??h ?h?? ?? ? ??k ???

  • WallTurb FP05/XZ - SPIV analysis (shell) (Application) [achieved]
    Description: WallTurb FP05/XZ
    TST04n5
    TST06

    Results: to do
    WallTurb/FP05/XZ
    Run Grid SPIV Rate
    Node CPU du (zlib)grid nb field time data (Mo/s) SPIV (3C/s)
    totalper node totalper node totalper node totalper CPU totalper CPU
    TST04n05 ?? ??? 18+71Go 1? Go ???x??? 1000x40 40x? ??h ?h?? ?? ? ??k ???
    TST06 ?? ??? 5.5 Go 5.5 Go ???x??? 15?? 5?x2? ??h ?h?? ?? ? ??k ???

  • WallTurb FP10/XZ - SPIV analysis (shell) (Application) [achieved]
    Description: WallTurb FP10/XZ
    TST04
    TST05

    Results: to do
    WallTurb/FP10/XZ
    Run Grid SPIV Rate
    Node CPU du (zlib)grid nb field time data (Mo/s) SPIV (3C/s)
    totalper node totalper node totalper node totalper CPU totalper CPU
    TST04 ?? ??? 20 Go ? Go ???x??? 1000x40 40x? ??h ?h?? ?? ? ??k ???
    TST05 ?? ??? 3.4 Go ? Go ???x??? 15?? 5?x2? ??h ?h?? ?? ? ??k ???

  • CALINS - SPIV analysis (shell) (Application) [achieved]
    Description: Run_000-203
    Run_900-901

    Results: ToDo
  • Monash/may2007 - SPIV analysis (shell) (Application) [in progress]
    Description:
    Part computed: Block 12 to 27:
    320 GB of uncompressed raw data (.b16)
    1936 2D3C fiels of 661x187 nodes

    Results:
    Monash/acquisition-may
    Run Grid SPIV Rate
    Node CPU du (raw)grid nb field time data (Mo/s) SPIV (3C/s)
    totalper node totalper node totalper node totalper CPU totalper CPU
    12-27 32 64 320Go 20 Go 661x187 1936 121/2 96h 1h30 60 4 44k 700

Publications

  • HR SPIV for dynamical system construction. [2009] (international)
    EntryType: inproceedings
    Author: J M Foucaut and S Coudert and M Stanislas
    Booktitle: Progress in Wall Turbulence: Understanding and modelling
    Month: {21 } # april
  • Spatial correlation from the SPIV database of the WALLTURB experiment. [2009] (international)
    EntryType: inproceedings
    Author: J M Foucaut and S Coudert and M Stanislas and J Delville
    Booktitle: Progress in Wall Turbulence: Understanding and modelling
    Month: {21 } # april
  • Study of the vortical structures in turbulent near wall flow. [2009] (international)
    EntryType: inproceedings
    Author: S Herpin and S Coudert and J M Foucaut and J Soria and M tanislas
    Booktitle: Progress in Wall Turbulence: Understanding and modelling
    Month: {21 } # april
  • Thick and Thin Volume Measurements of a Turbulent Boundary Layer using Tomographic Particle Image Velocimetry [2009] (international)
    EntryType: inproceedings
    Author: C. Atkinson, S. Coudert, J.M. Foucaut, M. Stanislas, J. Soria
    Booktitle: 8th {I}nternational {S}ymposium on {P}article {I}mage {V}elocimetry
    Month: 25-28 august
  • Unsteady characteristics of near-wall turbulence using high repetition stereoscopic particle image velocimetry (PIV) [2009] (international)
    EntryType: article
    Author: J M Foucaut and S Coudert and M Stanislas
    Journal: Measurement Science and Technology
    Volume: 20
    Number: 7
    Pages: 074004 (12pp)
    Url: http://stacks.iop.org/0957-0233/20/074004
    Abstract: This study is part of a project that is aimed at building dynamic boundary conditions near a solid wall, in order to reduce the large eddy simulation spatial resolution that is necessary in this region. The objective is to build a low-order dynamical system in a plane parallel to the wall, which will mimic the unsteady behaviour of turbulence. This dynamical system should be derived from a POD decomposition of the velocity field. The POD decomposition is to be applied on an experimental database of time-resolved velocity fields. In order to obtain the experimental database, a specific experiment of high-speed stereoscopic particle image velocimetry (PIV) has been performed. This experiment was carried out in the turbulent boundary layer of the LML wind tunnel. The plane under study was parallel to the wall located at 100 wall units. This database is validated via comparison with hot-wire anemometry (HWA). Despite some peak locking observed on the streamwise velocity component, the PDF and the power spectra are in very good agreement with the HWA results. The two-point spatial correlations are also in good agreement with the results from the literature. As the flow is time-resolved, space-time correlations are also computed. The convection of the flow structure is observed to be the most important effect at this wall distance. The next step is to compute the dynamical system and to couple it to a large eddy simulation.
  • Double large field Stereoscopic PIV in a high Reynolds number turbulent boundary layer [2007] (international)
    EntryType: inproceedings
    Author: S. COUDERT, J.M. FOUCAUT, J. KOSTAS
    Booktitle: 7th {I}nternational {S}ymposium on {P}article {I}mage {V}elocimetry
    Note: S. COUDERT, J.M. FOUCAUT, J. KOSTAS, M. STANISLAS Laboratoire de M\'ecanique de Lille (France) P. BRAUD, C. FOURMENT, J. DELVILLE Laboratoire d'Etudes A\'erodynamiques, Poitiers (France) M. TUTKUN, F. MEHDI, P. JOHANSSON, B. GEORGE Chalmers University of Technology (Sweden)
    Month: {11 } # sep
  • High spatial resolution stereoscopic DPIV in a turbulent boundary layer [2007] (international)
    EntryType: inproceedings
    Author: {Sophie Herpin}, Chong Y. Wong
    Booktitle: 7th {I}nternational {S}ymposium on {P}article {I}mage {V}elocimetry
    Note: Sophie Herpin1/2, Chong Y. Wong1, M. STANISLAS2, Julio Soria1 1 Department of Mechanical Engineering, Monash University, Clayton Campus, Victoria 3800, Australia 2 Laboratoire de Mecanique de Lille, Bd Paul Langevin, Cite Scientifique, Lille 59655, France
    Month: {11 } # sep
  • Study of the unsteady aspects of turbulence in the near wall region of a boundary layer using High Speed SPIV. [2007] (international)
    EntryType: inproceedings
    Author: J.M. FOUCAUT, S. COUDERT, J. KOSTAS
    Booktitle: 7th {I}nternational {S}ymposium on {P}article {I}mage {V}elocimetry
    Note: J.M. FOUCAUT, S. COUDERT, J. KOSTAS, M. STANISLAS Laboratoire de M\'ecanique de Lille (France) P. BRAUD, C. FOURMENT, J. DELVILLE Laboratoire d'Etudes A\'erodynamiques, Poitiers (France) M. TUTKUN, F. MEHDI, P. JOHANSSON, B. GEORGE Chalmers University of Technology (Sweden)
    Month: {11 } # sep

Collaborations

    Success stories and benefits from Grid'5000

    last update: 2011-06-22 11:20:00

    Personal tools
    Namespaces

    Variants
    Views
    Actions
    Public Portal
    Users Portal
    Admin portal
    Wiki special pages
    Toolbox