BigDFT

BigDFT is a DFT massively parallel electronic structure code using a wavelet basis set with the capability to use a linear scaling method. Wavelets form a real space basis set distributed on an adaptive mesh (two levels of resolution in our implementation). GTH or HGH pseudopotentials are used to remove the core electrons.

BigDFT is available in ABINIT v5.5 and higher but can also be downloaded in a standalone version from the website. Thanks to the developers’ Poisson solver based on a Green function formalism, periodic systems, surfaces and isolated systems can be simulated with explicit boundary conditions. The Poisson solver can also be downloaded and used independently and is integrated in ABINIT, Octopus and CP2K.

The code is free software, available under GNU-GPL license and the BigDFT developer community encourages anyone willing to contribute to join the team. The code, tutorials, and documentation are available on BigDFT site.

System requirements

Before running the NGC BigDFT container please ensure your system meets the following requirements.

• Pascal(sm60) or Volta(sm70) NVIDIA GPU(s)
• CUDA driver version >= 384.81
• One of the following container runtimes
  • nvidia-docker
  • Singularity >= v2.5.0

Running BigDFT

Command invocation

BigDFT generally doesn't take any executable arguments. Instead, input files are taken from the current working directory implicitly, specified by name. Output can be redirected as usual, using standard UNIX or mpirun mechanisms.

$ mpirun -np {num_procs} bigdft > log

Examples

The following examples demonstrate how to run the NGC BigDFT container under supported container runtimes.

• nvidia-docker
  • command line execution
  • interactive shell
  • jupiter notebook
  • mpi/openmp,gpu
• Singularity
  • command line execution
  • interactive shell
  • jupiter notebook
  • mpi,openmp,gpu

Running with nvidia-docker

The following modes of running are supported under nvidia-docker:

• command line execution
• interactive shell
• Jupyter notebook
• MPI, OpenMP, GPU

nvidia-docker aliases

DOCKER will be used to launch processes within the NGC BigDFT container using the nvidia-docker runtime:

$ export DOCKER="nvidia-docker run -it --rm -v $(pwd):/host_pwd -w /host_pwd nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl"

Where:

• DOCKER: alias used to store the base Docker command
• run: specifies the mode of execution
• -it: runs the container in an interactive tty shell
• --rm: makes the container instance ephemeral (does not save on exit)
• -v $(pwd):/host_pwd: bind mounts the current working directory into the container at /host_pwd
• -w /host_pwd: sets the inital working directory in the container to /host/pwd
• nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl: URI of the latest NGC BigDFT container

Keep in mind that DOCKER will be set to bind mount

Command line execution with nvidia-docker

To run the BigDFT container from the command line, mount the desired input files into the container. For example, to mount the current working directory into the container at /host_pwd/ and run BigDFT there:

$ nvidia-docker run -it --rm -v $(pwd):/host_pwd -w /host_pwd nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl bigdft

Where:

• run: specifies the mode of execution
• -it: runs the container in an interactive tty shell
• --rm: makes the container instance ephemeral (does not save on exit)
• -v $(pwd):/host_pwd: bind mounts the current working directory into the container at /host_pwd
• -w /host_pwd: sets the inital working directory in the container to /host/pwd
• nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl: URI of the latest NGC BigDFT container
• bigdft: the BigDFT executable

Note that the current working directory of the host must have read/write access for other in order for the container to access it via /host_pwd. Use chmod o+rw . to grant other access to the current directory

BigDFT requires input data to be present in the current directory. Example data is available within the container; to copy the FeHyb example data into the host's current directory, use:

$ nvidia-docker run -it --rm -v $(pwd):/host_pwd -w /host_pwd nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl cp -r /docker/FeHyb/GPU /host_pwd
$ cd GPU

Interactive shell with nvidia-docker

To run the container interactively, execute /bin/bash with the container:

$ nvidia-docker run -it --rm -v $(pwd):/host_pwd -w /host_pwd nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl /bin/bash

Where:

• run: specifies the mode of execution
• -it: runs the container in an interactive tty shell
• --rm: makes the container instance ephemeral (does not save on exit)
• -v $(pwd):/host_pwd: bind mounts the current working directory into the container at /host_pwd
• -w /host_pwd: sets the inital working directory in the container to /host/pwd
• nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl: URI of the latest NGC BigDFT container
• /bin/bash: starts an interactive Bash shell

To run BigDFT from the interactive shell, change to the directory containing the desired input data and run:

$ bigdft

Recall that the DOCKER alias binds the host's present working directory to /host_pwd when the container starts, which enables input from and output to the host's filesystem. Note that the current working directory of the host must have read/write access for other in order for the container to access it via /host_pwd. Use chmod o+rw . to grant other access to the current directory

Sample input data is also available within the container, including FeHyb data at /ContainerXp/FeHyb/GPU.

After the computation, output can be found in the log.yaml file, and timings in time.yaml. These files can be copied to /host_pwd for availability on the host after exiting the container.

Jupyter notebook with nvidia-docker

The NGC BigDFT container can also be controlled remotely via an interactive Jupyter interface. To start the Jupyter server, simply invoke the container with nvidia-docker:

nvidia-docker run -p 8888:8888 -it --rm -v $(pwd):/results nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl

This starts a Jupyter web interface on port 8888.

The default password of the Jupyter web interface is bigdft.

More documentation can be found on the BigDFT Documentation Webpage.

mpirun with nvidia-docker

OpenMPI is available within the NGC BigDFT container for multi-GPU utilization.

In order to run the NGC BigDFT container with OpenMPI, use:

$ nvidia-docker run -it --rm -v $(pwd):/host_pwd -w /host_pwd --ipc=host nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl mpirun -n <n> bigdft

Where:

• run: specifies the mode of execution
• -it: allows the container to use the host network devices (necessary to connect to the Kipoi API)
• --rm: makes the container instance ephemeral (does not save on exit)
• -v $(pwd):/host_pwd: bind mounts the current working directory into the container at /host_pwd
• -w /host_pwd: sets the inital working directory in the container to /host/pwd
• --ipc=host: sets the inter-process communication method
• nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl: URI of the latest NGC BigDFT container
• -n <n>: sets the number of MPI processes to <n>

It is recommended to set the number of MPI processes, <n>, equal to the number of GPUs available on the host.

Running with singularity

Pull the image

Save the NGC BigDFT container as a local Singularity image file:

$ singularity build bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg docker://nvcr.io/hpc/bigdft:cuda10-ubuntu1804-ompi4-mkl

This command saves the container in the current directory as bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg

Note: Singularity/2.x

In order to pull NGC images with singularity version 2.x and earlier, NGC container registry authentication credentials are required.

To set your NGC container registry authentication credentials:

$ export SINGULARITY_DOCKER_USERNAME='$oauthtoken'
$ export SINGULARITY_DOCKER_PASSWORD=<NVIDIA NGC Cloud Services API key>

More information describing how to obtain and use your NVIDIA NGC Cloud Services API key can be found here.

Local workstation with Singularity

This mode of running is suitable for interactive execution from a local workstation containing one or more GPUs. There are no requirements other than those stated in the System Requirements section.

Important Note for Amazon Machine Image users:

Amazon Machine Images on Amazon Web Service have a default root umask of 077. Singularity must be installed with a umask of 022 to run properly. To (re)install Singularity with correct permissions:

• Uninstall Singularity (if it is installed)
• Change the umask with: $ umask 0022
• Install Singularity
• Restore the umask: $ umask 0077

This causes installed Singularity files to have permission 0755 instead of the default 0700. Note that the umask command only applies changes to the current shell. Use umask and install Singularity from the same shell session.

Command line execution with Singularity

To run the BigDFT container from the command line, mount the desired input files into the container. For example, to mount the current working directory into the container at /host_pwd/ and run BigDFT there:

$ singularity run --nv -B $(pwd):/host_pwd --pwd /host_pwd bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg bigdft

Where:

• exec: specifies the mode of execution
• --nv: exposes the host GPUs to the container
• -B $(pwd):/host_pwd: bind mounts the current working directory into the container at /host_pwd
• --pwd /host_pwd: sets the working directory to /host_pwd when the container starts
• bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg: path of the saved Singularity image

The above command requires input data to be present in the host's current directory. Example data is available within the container; to copy the FeHyb example data into the host's current directory, use:

$ singularity exec --nv -B $(pwd):/host_pwd --pwd /host_pwd bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg /bin/bash -c "cp -r /docker/FeHyb/GPU /host_pwd"

Interactive shell with singularity

To run the container interactively, execute /bin/bash with the container:

$ singularity exec --nv -B $(pwd):/host_pwd --pwd /host_pwd bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg /bin/bash

Where:

• exec: specifies the mode of execution
• --nv: exposes the host GPUs to the container
• -B $(pwd):/host_pwd: bind mounts the current working directory into the container at /host_pwd
• --pwd /host_pwd: sets the working directory to /host_pwd when the container starts
• bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg: path of the saved Singularity image
• /bin/bash: starts an interactive Bash shell

To run BigDFT from the interactive shell, change to the directory containing the desired input data and run:

$ bigdft

Recall that the -B $(pwd:/host_pwd argument binds the host's present working directory to /host_pwd when the container starts, which enables input from and output to the host's filesystem.

Sample input data is also available within the container, including FeHyb data at /ContainerXp/FeHyb/GPU.

After the computation, output can be found in the log.yaml file, and timings in time.yaml. These files can be copied to /host_pwd for availability on the host after exiting the container.

Jupyter notebook with singularity

The NGC BigDFT container can also be controlled remotely via an interactive Jupyter interface. To start the Jupyter server, simply invoke the container with singularity run:

$ singularity run --nv -B $(pwd):/host_pwd --pwd /host_pwd bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg

This starts a Jupyter web interface on port 8888.

The default password of the Jupyter web interface is bigdft.

More documentation can be found on the BigDFT Documentation Webpage.

mpirun with singularity

OpenMPI is available within the NGC BigDFT container for multi-GPU utilization.

In order to run the NGC BigDFT container with OpenMPI, use:

$ singularity exec --nv -B $(pwd):/host_pwd --pwd /host_pwd bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg mpirun -n <n> bigdft

Where:

• exec: specifies the mode of execution
• --nv: exposes the host GPUs to the container
• -B $(pwd):/host_pwd: bind mounts the current working directory into the container at /host_pwd
• --pwd /host_pwd: sets the working directory to /host_pwd when the container starts
• bigdft_cuda9-ubuntu1604-mvapich2_gdr-mkl.simg: path of the saved Singularity image
• -n <n>: sets the number of MPI processes to <n>

It is recommended to set the number of MPI processes, <n>, equal to the number of GPUs available on the host.

Cluster mpirun with Singularity

Clusters with a local compatible OpenMPI installation may launch the NGC BigDFT container using the host provided mpirun or mpiexec launcher.

Cluster mpirun requirements

To use the cluster provided mpirun command to launch the NGC BigDFT container, OpenMPI/3.0.2 or newer is required.

Running with mpirun

Running with mpirun maintains tight integration with the resource manager.

Launch BigDFT within the container, using mpirun:

$ mpirun -n <N> --map-by ppr:<num_proc>:socket singularity exec --nv -B $(pwd):/host_pwd --pwd /host_pwd bigdft.simg

Where:

• <N>: MPI process count
• --map-by ppr:<num_proc>:socket: Distributes <num_proc> MPI ranks to each CPU socket
• exec: specifies the mode of execution

It is recommended to set the number of MPI processes, <N>, equal to the number of GPUs available, and <num_proc> to the number of GPUs with affinity to each CPU socket.

Container mpirun with Singularity

The NGC BigDFT container allows the user to launch parallel MPI jobs from fully within the container. This mode has the least host requirements, but does necessitate additional setup steps, as described below.

Singularity mpirun Requirements

• Passwordless rsh/ssh between compute nodes

Running mpirun in your container

The internal container OpenMPI installation requires an OpenMPI hostfile to specify the addresses of all nodes in the cluster. The OpenMPI hostfile takes the following form:

<hostname_1>
<hostname_2>
...
<hostname_n>

Generation of this nodelist file via bash script will vary from cluster to cluster. Common examples include:

SLURM

HOSTFILE=".hostfile.${SLURM_JOB_ID}"
for host in $(scontrol show hostnames); do
  echo "${host}" >> ${HOSTFILE}
done

PBS

HOSTFILE=$(pwd)/.hostfile.${PBS_JOBID}
for host in $(uniq ${PBS_NODEFILE}); do
  echo "${host}" >> ${HOSTFILE}
done

Additionally, mpirun must be configured to start the OpenMPI orted process within the container runtime. Set environment variables so that mpirun starts the OpenMPI orted process within the container:

$ export SIMG=bigdft.simg
$ export OMPI_MCA_plm=rsh
$ export OMPI_MCA_plm_rsh_args='-o UserKnownHostsFile=/dev/null -o StrictHostKeyChecking=no -o LogLevel=ERROR'
$ export OMPI_MCA_orte_launch_agent="${SINGULARITY} /usr/bin/orted"

To launch BigDFT using mpirun:

$ singularity exec --nv -B $(pwd):/host_pwd --pwd /host_pwd bigdft.simg mpirun --hostfile <hostfile> -n <N> --map-by ppr:<num_proc>:socket bigdft

Where:

• SINGULARITY: Singularity alias defined above
• <hostfile>: textfile list of compute hosts
• <N>: MPI process count
• --map-by ppr:<num_proc>:socket: Distributes <num_proc> MPI ranks to each CPU socket

Benchmarks

BigDFT includes two examples, each with CPU and GPU variants. These examples can be run with the instructions provided above.

• /ContainerXp/FeHyb/NOGPU : Directory containing CPU only FeHyb example
• /ContainerXp/FeHyb/GPU : Directory containing GPU accelerated FeHyb example
• /ContainerXp/H2O-32/CPU : Directory containing CPU only H2O example
• /ContainerXp/H2O-32/GPU : Directory containing GPU accelerated H2O example

The FeHyb CPU example includes an additional reference log file, log.ref.yaml, which can be used to check the correctness of theFeHyb output. To check the FeHyb output log.yaml, run the following command from an interactive shell within the container:

python /usr/local/bigdft/lib/python2.7/site-packages/fldiff_yaml.py -d /path/to/log.yaml -r /docker/FeHyb/NOGPU/log.ref.yaml -t /docker/FeHyb/NOGPU/tols-BigDFT.yaml

Where:

• /usr/local/bigdft/lib/python2.7/site-packages/fldiff_yaml.py: the path of the yaml file comparison tool
• -d /path/to/log.yaml: specifies the path to a FeHyb output log.yaml data file to check
• -r /docker/FeHyb/NOGPU/log.ref.yaml: path to the reference FeHyb output file
• -t /docker/FeHyb/NOGPU/tols-BigDFT.yaml: path to the FeHyb comparison tolerances file

The correct output should include:

Test succeeded: True

Suggested Reading

BigDFT documentation.

The BigDFT Python interface is documented here.