Introduction Julia

Welcome page and syllabus

  • Also link at House symbol 🏠 at top of page

Learning outcomes

  • Learn how to load Julia modules and access site-installed Julia packages

  • Grasp the concept of Julia environments

  • Familiarize with Julia packages installation with Pkg

  • Acquire the knowledge of writing batch scripts for running Julia

  • Discover and explore the different Julia parallelization tools

  • Learn how Julia supports GPU computing

  • Gain knowledge on how to run ML jobs in Julia

Your expectations?

  • Find best practices for using Julia at UPPMAX and HPC2N

  • Get the hang of packages functioning

  • Use the HPC performance with Julia

Not covered

  • Improve Julia coding skills

  • Other clusters

Julia Exercise files

  • On HPC2N, you can copy the Julia exercises in a tarball from /proj/nobackup/r-py-jl-m/exercises.tar.gz

  • On UPPMAX you can copy the exercises in a tarball from /proj/r-py-jl-m-rackham/exercises.tar.gz

  • On LUNARC you can copy the exercises in a tarball from ~bbrydsoe/exercises.tar.gz

Exercises

cd /proj/r-py-jl-m-rackham/<your-name>        # on Rackham
cd /proj/nobackup/r-py-jl-m/<your-name>       # on Kebnekaise
cp ../exercises.tar.gz .
tar zxvf exercises.tar.gz
tree exercises/julia

Preliminary schedule

Preliminary times

Time

Topic

Teacher

9:00

Log in

BC, PO

9.45

Coffee break

10:00

Syllabus

BC

10.15

Introduction, Julia in general

PO

10.25

Loading modules and running Julia codes

BC

10.55

Break

11.10

Packages and isolated environments

PO

12.00

LUNCH

13.00

SLURM Batch scripts for Julia jobs

PO

13.50

break

14.05

Parallel and multithreaded functions

PO

14.50

break

15.00

Simultaneous session — UPPMAX: Jupyter and interactive work

BC

.

Simultaneous session — HPC2N: Jupyter

PO

15.25

Summary

BC

15.30

Evaluation

15.40

Q&A on-demand

16.00

END

Instructor note

  • Intro 5 min

  • Lecture and 10 min

What is Julia?

Julia is a relatively new Programming language (pre-released 1.0 announced in 2018), compared to well-known and standard languages such as Fortran, C, C++, R, and Python (some of them dated back to the 70’s). A common pattern in those well-established languages (traditional paradigm) is that they look after either:

  • productivity (fast deployment, fast prototyping) or,

  • performance (fast numerical computations).

This pattern created the so called two-language problem where programmers had to choose between productivity (R, Python) or performance (Fortran, C/C++) and when both were needed programmers had to learn the language that offered the desired capability and use some interface between different languages. Performant languages traditionally need to be compiled while languages focusing on productivity are interpreted.

Among the different features of Julia, is its capability of integrating both aspects productivity and performance into a single language. In this way, Julia programmers can in principle write software without changing their focus to learn a new language.

../_images/two-language.png

Two-language problem where Julia is shown as a bridge between the languages in the traditional paradigms, productivity vs. performance.

Features of Julia

According to the Julia documentation some of the features of this language are:

  • Good Base library with efficiently implemented operations written in Julia itself

  • “Good performance, approaching that of statically-compiled languages like C”

  • Modular and self-contained approach for libraries and development

  • “A rich language of types for constructing and describing objects”

  • A fast growing community of users and developers

Shortcomings of Julia

  • As this is a new language, the libraries ecosystem is not as rich as in Python or R, for instance

  • Currently, using Julia for simple tasks (for instance, opening a file and writing text, plotting) is not as efficient as using Linux tools (AWK, GREP) or compiled languages (Fortran, C/C++, Rust)

  • Previous situation is more noticeable upon running simple tasks in parallel mode (MPI, Threads)

  • An initial code version can be fast (compared to base Python) with a code that is clear to novices and without spending a long time writing. However, if one needs to get a more optimized code, it would most likely increase its complexity (readability) and one would need to spend more time (learning/programming) as in the case of C/C++/Fortran.

More on Julia

Material for improving your programming skills

First level

The Carpentries teaches basic lab skills for research computing.

Second level

CodeRefinery develops and maintains training material on software best practices for researchers that already write code. Their material addresses all academic disciplines and tries to be as programming language-independent as possible.

  • Not yet anything Julia specific

ENCCS (EuroCC National Competence Centre Sweden) is a national centre that supports industry, public administration and academia accessing and using European supercomputers. They give higher-level training of programming and specific software.

Other NAISS centres

Keypoints

  • Julia is a relatively new language with several attractive features.

  • Julia purpose is to avoid changing between high performance and high productivity languages in the different phases of code development.