TCLB for Magnus

Scripts for compilation and running of TCLB at Magnus cluster

Introduction

Magnus cluster uses SLURM queue. So, to see queue you use squeue, to run jobs you run sbatch.

Computing nodes have different architecture then the login node - so nothing compiled for computational nodes can be runned on the login node. That why you need to submit a job for more or less anything you want to do. This includes configuration and compilation.

To ease the work with TCLB on Magnus, this repository of cluster-specific tools was created. If you clone it to p subdirectory, you can use it to compile and run TCLB with the below instructions.

Installation

Getting the source

git clone https://github.com/CFD-GO/TCLB.git
cd TCLB
git clone https://github.com/llaniewski/TCLB-Magnus.git p

Configuration

All the needed configuration and installation of R packages is done by:

p/config

Compilation

As R can be only executed on computational nodes, you cannot run make localy. Compilation (on 24 cores) can be executed using:

p/make d2q9

Running cases

To run cases you can use the run script:

p/run [model] [case.xml] [number of cores] [options for sbatch (optional)]

You can use the script from a different directory then TCLB main dir, so for instance:

cd /scratch/blarbla/blarbla/some_important_research/
~/TCLB/p/run d2q9 mycase.xml 48

If I understand the accouting on Magnus it rarely makes sens to run on less then 24 cores, as they bill you for 24.

Changeing options

You can change the default time (1h), name, etc by supplying additional options for sbatch like:

~/TCLB/p/run d2q9 mycase.xml 48 -J "new name" --time=24:00:00

Notifications

The scripts support getting PushBullet notifications based on: https://github.com/llaniewski/my.prompt