Skip to content

Latest commit

 

History

History
253 lines (186 loc) · 10.4 KB

README.md

File metadata and controls

253 lines (186 loc) · 10.4 KB
Raw

Introduction

In order to avoid installation hassle, this step-by-step tutorial can be completed in a pre-prepared environment that is contained in a Vagrant virtual machine. In order to start the virtual machine (VM), some dependencies must be installed:

Windows

  • Vagrant
  • VirtualBox

Mac

  • Vagrant
  • VirtualBox

Linux

  • Vagrant
  • VirtualBox OR libvirt

The VM is started with the following procedure:

  1. On the command line, navigate to the httk_tutorial folder that contains the file called Vagrantfile.
  2. Run the command vagrant up.
    • If there is need to specify the VM provider, use the --provider flag, e.g. vagrant up --provider virtualbox or vagrant up --provider libvirt.
  3. Run the command vagrant ssh.

You should now be in the shell of the virtual machine.

Installation

Install httk

  1. git clone https://www.github.com/httk/httk.git

  2. The "devel" branch is the most up to date: cd httk && git checkout devel

  3. Add source ~/httk/init.shell to your .bashrc file. Doing so will make the httk-* command line scripts available in your terminal. The init.shell will also add httk's Python directory to your $PYTHONPATH, so it takes care of the Python installation.

    • Restart the terminal or run the command source ~/.bashrc to activate the changes.

Note

Alternative ways of installing httk:

  • Developer install:
    • cd ~/httk && pip install --editable . --user
  • Normal install:
    • pip install httk

Initial httk setup

  1. Run the command httk-setup.
  2. Follow the instructions on screen.

A global configuration directory .httk will be created in your home folder. The folder has the following structure:

.httk/
├── computers
├── config
├── keys
│   ├── key1.priv
│   └── key1.pub
└── tasks
  • The computers folder contains global configuration of the supercomputers. Typically one configures computers within the httk project, so this folder can stay empty.
  • The config file contains some global config information, such as user's name and email address.
  • The keys folder contains to identify the "owner" of the data, e.g. when data is uploaded to a central database.
  • The tasks folder remains empty most of the time.

Setting up a new httk project

  • cd project
  • Run the command httk-project-setup.
  • Follow the instruction on screen.

A new folder ht.project will be created in the project folder with the following structure:

ht.project/
├── config
├── keys
│   ├── key1.priv
│   ├── key1.pub
│   └── owner.pub
├── references
└── tags

Setting up the computing clusters

  1. In the project folder, run the command httk-computer-setup
  2. When asked about setting up a project computer, answer yes.
  3. The is asked to choose a template for the computer:
    • local is for testing, e.g. on your laptop.
    • local-slurm if the httk project is located on the same supercomputer we want to run the calculations on.
    • ssh-slurm is for the remote computing clusters.
  4. We will setup a remote cluster, so choose ssh-slurm.
  5. Follow the instructions on screen:
    • Remote hostname
      • Refers to the SSH hostname of the cluster, e.g. tetralith.nsc.liu.se
    • Username
      • Login username for the cluster, when you SSH into the cluster, e.g. x_abcde.
    • Directory on remote host to keep runs and httk files: [Httk-runs]
      • The path where to keep httk-related files, e.g. /proj/theophys/users/x_abcde/Httk-runs.
    • The command to run vasp
      • The command for executing VASP in the cluster, e.g. mpprun /software/sse/manual/vasp/5.4.4.16052018/nsc1/vasp.
    • Vasp pseudopotential path
      • The absolute path to the pseudopotential folder, e.g. /software/sse/manual/vasp/POTCARs/potpaw_PBE.54
    • Slurm project
      • The SLURM account whose CPU hours we want to consume, e.g. snic2021-X-XXX.

Once the computer setup is complete, a new folder called <computer-name> is created in ht.project/computers. The configuration options that we specified above will be written to a file in ht.project/computers/<computer-name>/config.

Note

Currently httk does not have a sophisticated way to define how or how much computing resources should be allocated. In the config file one can change the SLURM_NODES parameter to correspond to the number of nodes we want the VASP calculations to parallelize over.

Advanced computer setup

The basic config file can be extended by additional config files that must be named config.<queue>. These additional config files contain options that either extend or override the basic options defined in config. The <queue> extension in the config.<queue> file refers to a "queue" on the cluster. The main purpose of queues is to keep tasks from different queues separate, so that the tasks can be managed separately.

  • For example, if we want to be able to submit calculations using multiple SLURM accounts, we can create a config file called e.g. config.snic2021-Y-YYY that has the following content:
SLURM_PROJECT="snic2021-Y-YYY"
  • As another example, often it is useful to be able to send quick test calculations to the testing/development reservation/partition of a cluster. This can be done by creating a file called e.g. config.devel with appropriately defined parameters, for example:
SLURM_NODES="1"
SLURM_RESERVATION="devel"
SLURM_TIMEOUT="0-00:30:00"

The values of the parameters should be chosen in such a way that the calculation "fits" in the testing/development reservation/partition.

Note

In order to use the SLURM_RESERVATION="devel" option a modified version of httk is needed, because it is not (yet) implemented in the official httk git repository.

Tip

There is generally no need to run httk-computer-setup every time we create a new project. In the future, it will be easier to copy the ht.project folder from your old project into your new project and edit the relevant config options with a text editor, e.g. the project_name parameter in the ht.project/config file should be given the name of your new project.

Generating VASP input files

In a high-throughput scenario we want to generate the many input files in an automated way. With httk one would typically write a Python script to do it. In the project folder there is an example script that generates input files for binary metal-nitrides in B1 structure.

  • Run the command python generate_Runs.py

Once the script finishes running, a Runs folder has been created that contains the tasks, each in their own folder. The contents of the Runs folder should look like this:

Runs/
└── tetralith
    ├── ht.task.unassigned.AlN_B1.start.0.unclaimed.3.waitstart
    │   ├── ht_steps
    │   ├── INCAR.relax
    │   └── POSCAR
    └── ht.task.unassigned.TiN_B1.start.0.unclaimed.3.waitstart
        ├── ht_steps
        ├── INCAR.relax
        └── POSCAR

Tasks can be divided to be executed by multiple different queues, for example to balance the usage of CPU hours per project. Here, all tasks are just under one queue, tetralith.

Note

httk has a standard way of naming these task folders ht.task.XXX. Information about the job and its status is encoded in the folder name using the "." character as a delimiter.

Installing httk on the cluster

Before tasks can be executed on the cluster, we have to install httk there.

  • In the project folder, run the command httk-computer-install <computer-name>.

Sending tasks to the cluster

Tasks are sent to the cluster with the command:

  • httk-tasks-send-to-computer <computer-name>:<queue> <runs-folder>

Note

If we want to use the "default" queue, then we can omit the :<queue> part in the command and just run

  • httk-tasks-send-to-computer <computer-name> <runs-folder>

In our case we will use the Tetralith cluster and the default queue, so the command looks like:

  • httk-tasks-send-to-computer tetralith Runs/tetralith.

Note

The <queue> parameter will determine where in the cluster the tasks are placed. The location has the form

  • ${REMOTE_HTTK_DIR}/Runs/<queue>,

where the ${REMOTE_HTTK_DIR} variable is defined in the computer's config file.

Start the taskmanager

The taskmanager that manages and runs the tasks on the cluster is started with the command:

  • httk-tasks-start-taskmanager <computer-name>:<queue> NUMBER

The NUMBER argument refers to the number of taskmanagers that will be spawned and run concurrently. Each taskmanager allocates the number of nodes that is defined in the <computer-name>'s config file with the SLURM_NODES parameter.

Note

The NUMBER argument is optional, and if it is omitted, only one taskmanager will be spawned.

What happens on the cluster?

For more information, check httk's official documentation.

  1. The taskmanager that we started on the cluster scans through the folder structure looking for tasks to run, which are the "special" folders named ht.task.XXX.
  2. Each task folder has a file named ht_steps, which is called the runscript. This script controls the flow of the VASP calculation(s). In our example, the ht_steps is told to relax the geometry (based on the settings in INCAR.relax) and then finish.

Receiving calculations from the cluster

Once the calculations have finished, one can download the output files from the cluster with the following command:

  • httk-tasks-receive-from-computer <computer-name>:<queue>

Storing results in a (local) SQLite database

VASP output files are analyzed and data stored in an SQLite database by httk. The project folder contains an example Python script make_database.py that accomplishes this.

  • Run the script: python make_database.py

An SQLite file called example.sqlite will be produced.

One can verify that a functional database was created by running the example read_database.py Python script. The printout should look something like this:

Simple search: Show all results:
Formula: AlN       , total_energy =  -14.530
Formula: NTi       , total_energy =  -19.572

A more advanced search: Show only Al-containing results:
Formula: AlN       , total_energy =  -14.530

A more advanced search: Show only the result with the lowest total energy:
Formula: NTi       , total_energy =  -19.572

Advanced topics

  • Using a Python runscript instead of the bash version.