ROAR_workshop

tutorial on how to do some basic stuff with PSU's supercomputing system

PART 0: Just use the RStudio Server!

1. Go to https://portal2.aci.ics.psu.edu/pun/sys/dashboard/ and log in.

2. Under Interactive Apps, click RStudio Server

3. Under Allocation, click "open" or whatever else is available to you.

4. You can use a custom environment if you want, but you don't have to. We'll start without one.

   a) Creating and using a custom environment is useful for code reproducibility since you can keep the version of R the same and you don't have reinstall the same R packages over and over again.

   b) You may need to use software like "gcc" or "github" and you can find which versions are available to use here: https://www.icds.psu.edu/computing-services/software/

5. Select an R version (only matters if there's no custom environment).

6. Select 1 hour, 1 core, and 5gb of memory per core.

   a) You should only select more than one core if you want to parallelize something, which we will do later. For now we just need one.

   b) The amount of memory you choose should be the amount of memory you think you'll use. Unless you're working with a large dataset or fitting a very complex model you shouldn't need much memory.

7. Aaaaand you all know how to use RStudio.

PARTS 1-2: Create a pbs script to submit an R code to the cluster

  #PBS -A open   

Allocation- Yours could also be the stat department allocation, "drh20_a_g_sc_default". There are 10 nodes and 20 processors per node available on this allocation.

  #PBS -l nodes=1:ppn=3 

Nodes and processors per node. Ex: "#PBS -l nodes=1:ppn=19": Request 19 ppn to parallelize your code across 18 processors. Request 19 ppn because one parent node is needed.

  #PBS -l walltime=48:00:00 

Requested wall time, format is HOURS:MINUTES:SECONDS. You can't request more than 48 hours. Your code will get to the front of the line to be run faster if you request less time.

  #PBS -l pmem=5GB 

Memory per processor. Be careful not to request more memory per node than is allowed in your specified allocation. Mine allows up to 250 gb total, which I think is also true for the stat department allocation since "sc" indicates standard core -- same kind of allocation.

  #PBS -N ProjName 

Job Name

  #PBS -j oe 

Prints log file for completion and errors

  #PBS -m abe

Sends email when job aborts, begins, and ends

  #PBS -M [email protected] 

Email Address

  cd /storage/work/PSUID/whereIsYourFile 

Changes directory to where your R file is located

  Rscript test.R 

Runs the R file

PART 3.1: Create a yml file and a pbs script for creating conda environments.

1. Create a new text file and add dependencies (a.k.a. R packages) as shown in parallel_env.yml

a) In general, when adding dependencies, you can find the right name by searching the name of the R package in the search bar at: https://anaconda.org/conda-forge/

2. Make sure to save the file name as parallel_env.yml (the .yml part is most important, DO NOT write .R!)

PART 3.2: Create a pbs script for creating conda environments.

1. Create another new text file with specifications as shown in parallel_env.pbs. The new parts of the pbs file are:

   #!/bin/bash
   

Need this to enable creation and usage of conda environments.

   module load python/3.6.8

Need this to set up a conda environment. The appropriate version of python is specified at https://www.icds.psu.edu/computing-services/software/

   module load anaconda3/2021.05

Need this to set up a conda environment. The appropriate version of anaconda is specified at https://www.icds.psu.edu/computing-services/software/

   conda env create --file parallel_env.yml

create the conda environment with the name of your .yml file in the specified folder!

   conda deactivate

Deactivate your conda environment when you're done using it.

PART 4: Run a R script within a conda environment.

1. The new parts of the pbs file are:

   conda activate parallel_env 
   

   Activates the conda environment you want to run your code in.