JuliaDelft24

A four-day course that will take place at TU Delft in October 2024.

Trainer: Dr. Carsten Bauer

Course page: https://www.tudelft.nl/en/evenementen/2024/dcse/courses/introduction-to-julia-for-high-performance-computing

Schedule (tentative)

Preparing for the course

Preparing your laptop (click to unfold)

Install Julia 1.10

The simplest way to install Julia 1.10 is via juliaup. Run one of the following in a terminal.

Linux/macOS

curl -fsSL https://install.julialang.org | sh -s -- --yes --default-channel 1.10

Windows

winget install julia -s msstore

Important: Julia 1.11 has been released on Oct 8 but for the course we'll still use 1.10. Make sure that you have the correct Julia version! To do so with juliaup you can run the following commands:

juliaup add 1.10
juliaup default 1.10

Download workshop materials

The simplest way to download the workshop materials (this GitHub repository) is through Git.

git clone https://github.com/carstenbauer/JuliaDelft24

If you don't have Git, you can either install it or manually download the materials as a .zip archive instead.

Installing the Julia environment

Within the JuliaDelft24 directory (that you've cloned or downloaded above), run the following command:

julia install.jl

Remark: The installation might take a couple of minutes and a few GB of disk space (worst case: up to ~10 minutes and up to ~2.6 GB). The reason is that we also install binary dependencies (e.g. MPI) via Julia's Package manager to be as self-contained as possible. If you want to remove everything after the course, simply delete ~/.julia.

Update PATH environment variable

We will use mpiexecjl during the course, which - after the installation above - lies in ~/.julia/bin. To make it available everywhere, we need to add ~/.julia/bin to the PATH environment variable. On Linux/macOS, you can add the following line to your .bashrc (or whatever file gets automatically loaded by your shell):

export PATH=$HOME/.julia/bin:$PATH

I don't have Windows, and don't know how to do it there, but you should readily find instructions on Google.

Install Visual Studio Code (+ extensions)

• Download Visual Studio Code from https://code.visualstudio.com/download and install it.
• Afterwards, install the following two extensions (the linked pages should have "Install" buttons, respectively)
  • Remote - SSH extension
  • Julia extension

Install Jupyter Lab

Technically, you don't necessarily need Jupyter Lab, because Visual Studio Code can also open notebook files. However, I highly recommend that you install it:

• Installation instructions

If you don't manage to install Jupyter using the above link, you can also let Julia try to install it for you. Run julia --project within the JuliaDelft24 directory and then execute the following Julia commands:

using IJulia
IJulia.notebook()

Preparing your DelftBlue account (click to unfold)

Login

Anyone with a TU Delft NetID should be able to SSH to DelftBlue. You can use a terminal or Visual Studio Code to login to the cluster. If necessary, more details are available in the DelftBlue documentation here.

Terminal

Open a terminal and login to the cluster with the following command (where your replace <netid> by your NetID).

ssh <netid>@login.delftblue.tudelft.nl

VS Code

1. Open Visual Studio Code.
2. Press CTRL + SHIFT + P or CMD + SHIFT + P (opens a popup menu) and type and select Remote-SSH: Connect to Host....
3. When asked for it, input <netid>login.delftblue.tudelft.nl for the hostname (with <netid> replaced by your NetID) and press Enter.

After some time, you should have VS Code running on the cluster. You can get an integrated terminal by pressing CTRL + SHIFT + P or CMD + SHIFT + P and running Terminal: Create New Terminal.

Setting things up

Execute the following command on the cluster

sh /projects/julia/setup_account.sh

You only have to do this once. If you're curious, the script will

• put a single line at the end of your ~/.bashrc,
• clone the workshop materials to /scratch/<netid>/JuliaDelft24.

(Easy to undo after the course, if you like to.)