Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Repo info
Activity
    Niyazi Cem Degirmenci
    @ncde
    @victorsndvg perfect thank you
    Niyazi Cem Degirmenci
    @ncde
    Hello I started filling the benchmark document but now saw that for the one node case the gnu singularity is faster compared to gnu native
    and the results are an average of 10 runs
    do you think there is something different between gnu native and gnu singularity environments?
    victorsndvg
    @victorsndvg
    And what about the comparison of minimum obtained times?
    Niyazi Cem Degirmenci
    @ncde
    singularity beats in that too
    but just for gnu
    for intel everything is normal
    victorsndvg
    @victorsndvg
    There is no difference between the mirror container and the native system. The containers are built from a tar pipe containing the native OS
    which queue/partition you used?
    how you launched the benchmarking?
    Niyazi Cem Degirmenci
    @ncde
    for native gnu:
    '''

    '''#!/bin/bash

    SBATCH -t 00:20:00 # execution time. Ex:01:00:00

    SBATCH -n 20 # number of tasks. Ex:8

    SBATCH -p thinnodes

    module purge

    module load intel/2016 impi/5.1 fenics-hpc/0.8.4

    module load singularity/2.2.1

    module load gcc/5.3.0 impi/5.1 fenics-hpc/0.8.4

    for i in {1..10}
    do

       #srun -n 20 $SINGULARITY exec -B /etc:/etc -B /var:/var -B  /mnt:/mnt -B /scratch:/scratch --pwd $(pwd) $CONTAINER   ./poisson meshes/cube_level_5.bin 1> log_$i 2 > logerr_$i

    mpirun singularity exec ../../images/optimization_o2.img ./poisson meshes/cube_level5.bin > log$i

        mpirun -n 20 ./poisson meshes/cube_level_5.bin >log_$i

    done

    opps
    victorsndvg
    @victorsndvg
    Perfect, I understand. You used thinnodes partition
    in thinnodes you have exclusive usage of the resources
    Niyazi Cem Degirmenci
    @ncde
    should I use a different partition for benchmarking?
    victorsndvg
    @victorsndvg
    No, thinnodesis ok. Depending on the partition, you must specify --exclusive. but NOT inthinnodes`, is the default behaviour
    Niyazi Cem Degirmenci
    @ncde
    copying and pasting here doesn't work well
    some commented lines show bigger
    but these tests were done in /mnt/lustre/scratch/home/otras/cfd/nde/MSO4SC/tests-2017-06-07/single_node
    where I also keep the run10_xxx.batch files that I used for submitting
    victorsndvg
    @victorsndvg
    Ok, I'm going to take a look
    Tamás Budai
    @budait
    @ncde @victorsndvg I executed the cube demo in singularity and did the first post-processing step according to job-adaptive_ft2, however when I try to load the resulting pvd files into paraview 5.3, the is an error:
    Cannot read point data array "U" from PointData in piece 0. The data array in the element may be too short.
    Niyazi Cem Degirmenci
    @ncde
    can you try to read a vtu file?
    Tamás Budai
    @budait
    just opened the first vtu file, same error
    Niyazi Cem Degirmenci
    @ncde
    for me it works
    I created a postprocessing example at /mnt/lustre/scratch/home/otras/cfd/nde/MSO4SC/tests-2017-06-07/5_nodes/unicorn_gnu_nat/cube_sim01_postp
    where I used ./dolfin_post -m mesh_out.bin -t vtk -n 200 -s velocity -f 10 -t vtk
    second -t vtk is not necessary
    Tamás Budai
    @budait
    okay I will try that
    Niyazi Cem Degirmenci
    @ncde
    I have finished the performance simulations and updated the document https://docs.google.com/document/d/176skfyqcJYwK04moMgg_23K52u5_hY7JR1EUu7fgzA8/edit
    victorsndvg
    @victorsndvg
    Thanks @ncde :+1: great!
    I'm going to add a simple HPL benchmark with singularity at FT2 in order to close this sector
    Massimiliano Leoni
    @massimiliano-leoni
    @budait If you ran the simulation with just one process, you ran into a known issue of dolfin-hpc, see https://bitbucket.org/fenics-hpc/dolfin-hpc/issues/9/mesh-i-o-could-be-broken-in-serial .
    Tamás Budai
    @budait
    hi everyone! I would like to update a fenics test machine we have
    can I use singularity images or is better to compile dofin-hpc from source?
    victorsndvg
    @victorsndvg
    I think @ncde is the one in charge of FEniCS-HPC Singularity images
    johanhoffman
    @johanhoffman
    Yes, that is correct. Best to contact @ncde
    Tamás Budai
    @budait
    thanks!
    Niyazi Cem Degirmenci
    @ncde
    Hello I will provide an image to you Tamas, which application will you be running?
    @budait
    though the images that I can create on my laptop use openmpi and gcc as I am not superuser on the machines that have access to intelmpi and intel compilers
    Tamás Budai
    @budait
    Hi @ncde there is speicific application, just want to be able to run dolfin-hpc
    Niyazi Cem Degirmenci
    @ncde
    /mnt/lustre/scratch/newhome/otras/cfd/nde/orchestrator_tests/nautilus.simg is one image that you can use
    on FT2
    if you give me your user name on FT2 I can give you access to that folder
    Tamás Budai
    @budait
    I can create the image myself if i can get the receipt for it
    Andrei Berceanu
    @berceanu
    would you guys say fenics is a suitable alternative to comsol?