- A Finite Element Software Teresa Beck, Simon Gawlok and HiFlow - - PowerPoint PPT Presentation

a finite element software
SMART_READER_LITE
LIVE PREVIEW

- A Finite Element Software Teresa Beck, Simon Gawlok and HiFlow - - PowerPoint PPT Presentation

- A Finite Element Software - A Finite Element Software Teresa Beck, Simon Gawlok and HiFlow team HiFlow-Finite Element Software Introduction Introduction parallel finite element software developed by EMCL EMCL (Engineering


slide-1
SLIDE 1
  • A Finite Element Software
  • A Finite Element Software

Teresa Beck, Simon Gawlok and HiFlow³ team

slide-2
SLIDE 2

– parallel finite element software – developed by EMCL EMCL (Engineering Mathematics and Computing Lab)

  • f Prof. Heuveline, IWR, University of Heidelberg

– 12 years of development and experience – open source

  • pen source: LGPLv3-License

LGPLv3-License

HiFlow³-Finite Element Software

Introduction Introduction

slide-3
SLIDE 3

Mesh Mesh

  • 2D: triangles, quads
  • 3D: tetrahedrons,

hexahedrons

  • unstructured meshes
  • h-refinement

Finite Element Spaces Finite Element Spaces

  • Lagrange Finite Elements
  • arbitrary polynomial degree
  • p-refinement

User defined application User defined application

  • PDE
  • assembly of matrices and vectors
  • postprocessing
  • visualization

Linear Algebra toolbox Linear Algebra toolbox

  • matrix and vector structures
  • linear and nonlinear solvers
  • preconditioners

HiFlow³-Finite Element Software

A modular approach A modular approach

slide-4
SLIDE 4

– interfaces interfaces to various toolkits: METIS, MUMPS, ATLAS, MKL BLAS, HDF5, CLAPLACK, MKL LAPACK, CUDA, OpenCL, GaussQ, ILU++, OpenMP, UMFPACK – backends backends for matrix and vector node-level implementation: CUDA, OpenMP, naive, OpenCL, … – parallelism parallelism introduced on three levels: – distributed memory parallelization: MPI MPI – shared memory parallelization: OpenMP OpenMP – accelerators: CUDA, OpenCL CUDA, OpenCL

HiFlow³-Finite Element Software

Interfaces and Backends Interfaces and Backends

slide-5
SLIDE 5

HiFlow³-Finite Element Software

Performance and Scalability Performance and Scalability

Nozzle benchmark: Nozzle benchmark:

  • Incompressible Navier-Stokes equations
  • Reynolds number approx. 500
  • Discretization with P2/P1 elements
  • about 4 Mio. unknowns
  • Block preconditioning with ILU++
  • GMRES iterative linear solver
  • Newton method
slide-6
SLIDE 6

HiFlow³-Finite Element Software

Performance and Scalability Performance and Scalability

Scaling of Nozzle benchmark on JUROPA, FZ Jülich Scaling of Nozzle benchmark on JUROPA, FZ Jülich

slide-7
SLIDE 7

HiFlow³-Finite Element Software

Performance and Scalability Performance and Scalability

Scaling of Nozzle benchmark on JUQUEEN, FZ Jülich Scaling of Nozzle benchmark on JUQUEEN, FZ Jülich

slide-8
SLIDE 8

HiFlow³-Finite Element Software

Fields of Application Fields of Application

amongst others: amongst others:

Environmental Sciences

Baroclinic Baroclinic Wavetank Wavetank

Environmental Sciences

Tropical Tropical Cyclones Cyclones

Medical Engineering

Aortic Blood Aortic Blood Flow Flow

Biochemistry

Chromatographic Chromatographic System System

slide-9
SLIDE 9

HiFlow³-Finite Element Software

Goal Oriented Adaptivity Goal Oriented Adaptivity

For Tropical Cyclones For Tropical Cyclones

~100km ~10,000km ~1500km

?

Goal – prediction of storm tracks and intensity Challenges for the modeling – multi-scale problem – Which regions and which processes are relevant?

Approach: goal-oriented adaptivity in space and time Approach: goal-oriented adaptivity in space and time

Reference: Baumann, M., Numerical Simulation of Tropical Cyclones using Goal-Oriented Adaptivity, Phd-Thesis, Karlsruhe Institute of Technology, 2011

slide-10
SLIDE 10

HiFlow³-Finite Element Software

Goal Oriented Adaptivity Goal Oriented Adaptivity

For Tropical Cyclones For Tropical Cyclones

Implementation highlights with

– finite element discretizations in space/time (Q2-Q1/cGP(1)) – h-/Δt-adaptivity in space/time – adaptivity guided by goal-oriented error estimators – computation of dual solution – higher-order interpolation – mesh adaptation strategy – preconditioning with ILU++ – simulations with up to 10 Mio. unkowns in 3D

Reference: Baumann, M., Numerical Simulation of Tropical Cyclones using Goal-Oriented Adaptivity, Phd-Thesis, Karlsruhe Institute of Technology, 2011

slide-11
SLIDE 11

HiFlow³-Finite Element Software

Release 1.4 Release 1.4

– available from 11/07/2014 – maintenance release maintenance release – streamlined and standardized version – new features new features

– generic interfaces for Matrices and Vectors – additional postprocessing abilities – support for geometric search – evaluation of solution at arbitrary points – extended support for single precision

– two new tutorials two new tutorials – improved examples improved examples

slide-12
SLIDE 12

HiFlow³-Finite Element Software

Outlook Outlook

Release 1.5 Release 1.5

– available from autumn 2014 – new parallel I/O concept new parallel I/O concept

– based on XDMF XDMF (eXtensible Data Model and Format) and HDF5 HDF5 (hierarchival data format) – enables using the same parallel I/O data format for both visualization and checkpointing

– new module Stochastic FEM new module Stochastic FEM

– model uncertainties in physical problems – support of specialized solvers and preconditioners

Stay up to date and subscribe to our Newsletter! Stay up to date and subscribe to our Newsletter!

slide-13
SLIDE 13

HiFlow³-Finite Element Software

www.hiflow3.org www.hiflow3.org

– – Documentation Documentation – – Show cases Show cases – – Tutorials Tutorials – – Forum Forum – – Download Download – – ... ...

slide-14
SLIDE 14

HiFlow³-Finite Element Software

Thanks Thanks for for your your attention! attention!