See for additional information.


It is one of the critical software to use, e.g.,

module avail gcc
gcc --version


gfortran --version


The Glasgow Haskell Compiler is seen from module avail ghc, e.g.,

module load ghc/8.2.2
ghc --version


The popular git can be loaded,

git --help
git add --help


It is avail from /usr/bin/go and also visiable from module avail go.


module avail openjdk
java -version


The Julia compiler is visible from module avail julia, and by default it loads 1.6.2

module load gcc/9 julia
julia --version

Here is the "hello, world!" session,

   _       _ _(_)_     |  Documentation:
  (_)     | (_) (_)    |
   _ _   _| |_  __ _   |  Type "?" for help, "]?" for Pkg help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 1.6.2 (2021-07-14)
 _/ |\__'_|_|_|\__'_|  |
|__/                   |

julia> println("Hello World")
Hello World



Official website:

The executables (oocalc, ooffice, ooimpress, oomath, ooviewdoc, oowriter) are in the /usr/bin directory and can be conveniently called from the console, e.g.,

oowriter README.docx

to load the Word document.


Official website:

module avail matlab
module load matlab/r2019b

followed by matlab.


One could access databases elsewhere, e.g., at UCSC – see examples on VEP.

There isn't any MySQL cluster running as a general service on CSD3. Do you believe your group has something running on a VM hosted on our network possibly? If you need a database for your work, running it in your own department and then allowing access to it from CSD3. Databases are not suitable candidates to run on a HPC cluster, the resource requirements are different and by definition they need to be running continuously whilst access is required, so wouldn't be run via SLURM for example.


Official website:

module load ceuadmin/pspp

with command-line tool pspp and a GUI counterpart psppire.


Official website:

This can be invoked from a CSD3 console via python and python3. Libraries can be installed via pip and pip3 (or equivalently python -m pip install and python3 -m pip install), e.g., the script

pip install mygene --user
pip install tensorflow --user
pip install keras --user
pip install jupyter --user

installs libraries at $HOME/.local.

It is advised to use virual environments, i.e.,

# inherit system-wide packages as well
module load python/3.5
virtualenv --system-site-packages py35
source py35/bin/activate
# pip new packages

An alternative syntax is python3 -m venv py37

Note that when this is set up, one only needs to restart from the source command. The pip is appropriate for installing small number of package; otherwise Anaconda ( and Jupyter notebook ( are useful.

We first load Anaconda and create virtual environments,

module avail miniconda
module load miniconda/2
conda create -n py27 python=2.7 ipykernel
source activate py27

for Python 2.7 at /home/$USER/.conda/envs/py27, where envs could be replaced with the --prefix option. These are only required once.

We can also load Anaconda and activate Python 3.51,

module load miniconda/3
conda create -n py35 python=3.5 ipykernel
source activate py35

and follow Autoencoder in Keras tutorial on data from

The Jupyter notebook can be started as follows,

$HOME/.local/bin/jupyter notebook --ip= --no-browser --port 8081

If it fails to assign the port number, let the system choose (by dropping the --port option). The process which use the port can be shown with lsof -i:8081 or stopped by lsof -ti:8081 | xargs kill -9. The command hostname gives the name of the node. Once the port number is assigned, it is used by another ssh session elsewhere and the URL generated openable from a browser, e.g.,

ssh -4 -L 8081: -fN <hostname>
firefox <URL> &

paying attention to the port number as it may change.

An hello world example is hello.ipynb from which hello.html and hello.pdf were generated with jupyter nbconvert --to html|pdf hello.ipynb.

See also and

See HPC docuementation for additional information on PyTorch, Tensorflow and GPU.

:star: Introduction to HPC in Python (GitHub).


Official website: and also

Under HPC, the default version is 3.3.3 with /usr/bin/R; alternatively choose the desired version of R from

module avail R
module avail r
# if you would also like to use RStudio
module avail rstudio

e.g., module load r-3.6.0-gcc-5.4.0-bzuuksv rstudio/1.1.383.

For information about Bioconductor installation, see

The following code installs package for weighted correlation network analysis (WGCNA).

# from CRAN
dependencies <- c("matrixStats", "Hmisc", "splines", "foreach", "doParallel",
                  "fastcluster", "dynamicTreeCut", "survival")
# from Bioconductor
biocPackages <- c("GO.db", "preprocessCore", "impute", "AnnotationDbi")
# R < 3.5.0
# R >= 3.5.0

A good alternative is to use remotes or devtools package, e.g.,


A separate example is from r-forge, e.g.,

rforge <- ""
install.packages("estimate", repos=rforge, dependencies=TRUE)

In case of difficulty it is still useful to install directly, e.g.,

R CMD INSTALL ontoCAT_1.8.0.tar.gz
# Alternatively,
R -e "install.packages('ontoCAT_1.8.0.tar.gz',repos=NULL)"

The package installation directory can be spefied explicitly with R_LIBS, i.e.,

export R_LIBS=/rds/user/$USER/hpc-work/R:/rds/user/$USER/hpc-work/R-3.6.1/library

To upgrade Bioconductor, we can specify as follows,

BiocManager::install(version = "3.14")


module load ceuadmin/ruby
ruby --version


Official website:

Location at csd3: /usr/local/Cluster-Docs/SLURM/.

Account details


Note that after software updates on 26/4/2022, this command only works on non-login nodes such as icelake.


scontrol show partition

An interacive job

sintr -A MYPROJECT -p skylake -N2 -n2 -t 1:0:0 --qos=INTR

and also

srun -N1 -n1 -c4 -p skylake-himem -t 12:0:0 --pty bash -i

then check with squeue -u $USER, qstat -u $USER and sacct. The directory /usr/local/software/slurm/current/bin/ contains all the executables while sample scripts are in /usr/local/Cluster-Docs/SLURM, e.g., template for Skylake.

NOTE the skylakes are approaching end of life, see and For Ampere GPU, see

Start a job at a specific time

This is achieved with the -b or --begin option, e.g.,

sbatch --begin=now+3hour

Holding and releasing jobs

Suppose a job with id 59230836 is running, they can be achieved with,

scontrol hold 59230836
control release 59230836


Monitoring jobs

This is done with squeue command as above.

Use of modules

The following is part of a real implementation.

. /etc/profile.d/
module purge
module load rhel7/default-peta4
module load gcc/6
module load aria2-1.33.1-gcc-5.4.0-r36jubs

An example

To convert a large number of PDF files (INTERVAL.*.manhattn.pdf) to PNG with smaller file sizes. To start, we build a file list, and pipe into parallel.

ls *pdf | \
sed 's/INTERVAL.//g;s/.manhattan.pdf//g' | \
parallel -j8 -C' ' '
  echo {}
  pdftopng -r 300 INTERVAL.{}.manhattan.pdf
  mv {}-000001.png INTERVAL.{}.png

which is equivalent to SLURM implementation using job arrays (


#SBATCH --ntasks=1
#SBATCH --job-name=pdftopng
#SBATCH --time=6:00:00
#SBATCH --cpus-per-task=8
#SBATCH --partition=skylake
#SBATCH --array=1-50%10
#SBATCH --output=pdftopng_%A_%a.out
#SBATCH --error=pdftopng_%A_%a.err
#SBATCH --export ALL

export p=$(awk 'NR==ENVIRON["SLURM_ARRAY_TASK_ID"]' INTERVAL.list)
export TMPDIR=/rds/user/$USER/hpc-work/

echo ${p}
pdftopng -r 300 INTERVAL.${p}.manhattan.pdf ${p}
mv ${p}-000001.png INTERVAL.${p}.png

invoked by sbatch. As with Cardio, it is helpful to set a temporary directory, i.e.,

export TMPDIR=/rds/user/$USER/hpc-work/

Neither parallel nor SLURM

The following script moves all files a day earlier to directory old/,

find . -mtime +1 | xargs -l -I {} mv {} old

while the code below downloads the SCALLOP-cvd1 sumstats for proteins listed in cvd1.txt.

export url=
if [ ! -d ~/rds/results/public/proteomics/scallop-cvd1 ]; then mkdir ~/rds/results/public/proteomics/scallop-cvd1; fi
cat cvd1.txt | xargs -I {} bash -c "wget ${url}/{}.txt.gz -O ~/rds/results/public/proteomics/scallop-cvd1/{}.txt.gz"
#  ln -s ~/rds/results/public/proteomics/scallop-cvd1

Trouble shooting

With error message

squeue: error: _parse_next_key: Parsing error at unrecognized key:
squeue: error: Parse error in file
/usr/local/software/slurm/slurm-20.11.4/etc/slurm.conf line 22:
"InteractiveStepOptions="--pty --preserve-env --mpi=none $SHELL""
squeue: fatal: Unable to process configuration file

then either log out and login again, or



Official website:

As a CEU member the following is possible,

module load ceuadmin/stata/14

as with ceuadmin/stata/15. The meta-analysis (metan) and Mendelian Randomisation (mrrobust) packages can be installed as follows,

ssc install metan
net install mrrobust, from("") replace
  1. The following error

    CustomValidationError: Parameter channel_priority = 'flexible' declared in <<merged>> is invalid.
    The value 'flexible' cannot be boolified.

    can be resolved with conda config --set channel_priority false