Julia

Julia was created as an open source alternative to MATLAB and Octave.

External Links:
Julia website
Learning Julia


Loading Julia

There are various versions of Julia on Henry2. To see the various versions available, type module avail . To set the environment, load the module.

module load julia 

Running Julia

Do not use Julia on a login node. Submit Julia scripts with a batch submission file, and use the HPC-VCL if doing interactive visualization with Julia.

Batch Serial Julia Job

Here is an example batch script. For more information on submitting batch scripts, see the documentation on running jobs.

For a sample Julia script called my_code.jl containing

# a simple Julia code 
using LinearAlgebra

a = [1,2,4];
println("a",a)
b = normalize(a)
println("normalize(a)",b)
println("norm(b)",norm(b))

and a sample submission script submit.sh

#!/bin/bash
#BSUB -n 1        
#BSUB -W 30
#BSUB -o out.%J
#BSUB -e err.%J

module load julia
julia -p 1 my_code.jl

a batch job is submitted as

bsub < submit.sh

Script submit.sh requests the use of 1 core (-n 1) for up to 30 minutes (-W 30). The output in out.JOBID is

a[1, 2, 4]
normalize(a)[0.2182178902359924, 0.4364357804719848, 0.8728715609439696]
norm(b)1.0

Check the documentation on running jobs to customize the batch script.

Batch Shared Memory Parallelization (Multithreading) Julia Job

If your Julia script takes advantage of multithreading (i.e., using more than one core on just one node for parallelization), then you need to change your batch submission file accordingly.

Here is an example “Hello World” .jl script for multithreading called test_multithread.jl:

import Printf
N=4
Threads.@threads for i in 1:N
  sleep(1)
  Printf.@printf "Hello from thread %d\n" i
end   

Here is a sample submission script submit.sh

#!/bin/bash
#BSUB -n 4
#BSUB -W 30
#BSUB -R span[hosts=1]
#BSUB -o out.%J
#BSUB -e err.%J
    
module load julia
julia -t 4 test_multithread.jl
    

The batch job is submitted as

bsub < submit.sh

and the output in out.JOBID looks similar to (order of threads differs each time)

Hello from thread 1
Hello from thread 2
Hello from thread 4
Hello from thread 3    

Batch Distributed Memory Parallelization (MPI) Julia Job

If your Julia script takes advantage of MPI (i.e., using multiple cores on multiple nodes for parallelization), then you need to change your batch submission file accordingly.

However, before designing and submitting a batch job, there are some package installation procedures and MPI configuration steps that must take place:

1. Login to a Hazel login node
2. mkdir -p path/to/your/julia/packages(fill in correct path)
3. export JULIA_DEPOT_PATH=path/to/your/julia/packages(note, must be in a location that compute nodes have write access to . /share is not a great choice, because files not accessed in 30 days are deleted. /usr/local/usrapps/ < your group name > will not work b/c it is not writeable from compute nodes. The PI’s Research Storage is a good option: OIT Research Storage
4. module load julia(or the specific version of julia you are using)
5. module load openmpi-gcc
6. Julia
7. (now from within julia) import Pkg
8. Pkg.add("MPI")
9. Pkg.add("MPIPreferences")
10. exit()
11. (now back at login terminal shell)julia --project -e 'using MPIPreferences; MPIPreferences.use_system_binary()'

Here is an example “Hello World” .jl script for MPI: example

Here is a sample submission script submit.sh

#!/bin/bash
#BSUB -n 4
#BSUB -W 30
#BSUB -R span[ptile=2]
#BSUB -o out.%J
#BSUB -e err.%J
    
export JULIA_DEPOT_PATH=path/to/your/julia/packages  # file in correct path, the path must be writeable by a compute node (i.e., not /usr/local/usrapps/$GROUP), and it shoud not be $HOME either because it does not have enough space to install the packages
    
module load julia
module load openmpi-gcc/
    
mpirun julia 04-sendrecv.jl
    

A batch job is submitted as

bsub < submit.sh

and the output in out.JOBID looks similar to (order may be different, in fact, it may be different every time you run it)

1: Sending   1 -> 2 = [1.0, 1.0, 1.0, 1.0]
3: Sending   3 -> 0 = [3.0, 3.0, 3.0, 3.0]
0: Sending   0 -> 1 = [0.0, 0.0, 0.0, 0.0]
0: Received 3 -> 0 = [3.0, 3.0, 3.0, 3.0]
2: Sending   2 -> 3 = [2.0, 2.0, 2.0, 2.0]
3: Received 2 -> 3 = [2.0, 2.0, 2.0, 2.0]
2: Received 1 -> 2 = [1.0, 1.0, 1.0, 1.0]
1: Received 0 -> 1 = [0.0, 0.0, 0.0, 0.0]
    

Check the documentation on running jobs to customize the batch script.

Installing Julia packages

Additional packages outside of the base Julia install must be installed by the user. See the instructions for request a space for user maintained software.

The basic process to install a new Julia package is to use Pkg.add from a login node. By default, Julia will install the packages in the home directory in ~/.julia. That will quickly fill the home directory. Before installing packages, set the location for the installs by setting JULIA_DEPOT_PATH to the user maintained software directory.

export JULIA_DEPOT_PATH = /usr/local/usrapps/[your space for user maintained software]/[whatever you want to call the julia apps directory]

module load julia
julia
import Pkg
Pkg.add("whatever")

Check the Julia documentation on setting environment variables.

Last modified: April 12 2025 13:38:20.