01-08-2024 Vilnius Meeting Notes

S1: Round of intros

Everyone

• Collected names and emails of participants

S1: Review Finland Meeting

@nicolin

@Deepak Tunuguntla

• DEM-HPC

  • What about OpenCL? - Robertas Navakas

    • This is pretty outdated in terms of supporting newer platforms. CUDA is proprietory

    • This is where kokkos is helpful

• Brief introduction to discrete simulations

  • Simulation methods

    • event based + monte carlo

    • molecular dynamics

    • discrete element method

  • Neighbour search methods

• Machine learning

S1: Recent developments

@nicolin

@Deepak Tunuguntla

• Introducing the chosen benchmarking cases

  • academic

  • industrial

S2: Brief recap of S1 (pre-lunch session) and tasks defined

@Deepak Tunuguntla

@nicolin

• DEM-HPC

  • Neighbourhood search

    • Broad phase, mid phase, narrow phase

    • Task 1: list articles representing the current state of the art w.r.t contact detection

      • Start with reading this review https://www.sciencedirect.com/science/article/pii/S0266352X20302718#s0010

      • Anything missing please add them to a confluence page, which will be shared later

        • articles, scripts, libraries, repos

  • Time integration algorithms

    • Euler

    • Leap frog

    • Velocity verlet

    • Runge-Kutta

    • Position Verlet

    • Predictor corrector

    • Things w.r.t time-step, precision issues

    • Task 2: Compare performance of own/open/closed source codes for simple and benchmarking cases

      • OS: Yade, liggghts, MercuryDPM

        • Yade: Radan Ivanov (BG), Aleksander (Greece)

        • liggghts: Robertas (LT)

        • MercuryDPM: @Deepak Tunuguntla (NL)

      • Own codes:

        • XDEM - Bernhard Peters(Lux)

        • - Yannick Descantes (Fra)

      • Closed/Commercial codes

        • abacus - Francisco Gilabert (Bel)

        • BlazeDEM - @nicolin (UK), Rafal (Pol)

        • ScaleDEM - Francisco Gilabert (Bel)

  • Output format

    • List your code output formats in the confluence page we will share after this meeting

  • Hardware

    • Start with comparing performance on a CPU

  • Visualisation

    • Stick to paraview

      • For spheres, script to output in a standard paraview settings

  • Parallel computing libraries

    • OpenMP, MPI, kokkos

  • Other simulation methods

    • Event-based

      • impuls (contact dynamics), LMCG, bullet, chronos

      • Is a hybrid solver possible?

  • Programming languages : C++ and Python