AAA- Quick Start Guide

Logging in:

To login to the clusters, the easiest way is with 'ssh'.  On Windows, open a PowerShell - on the Mac, open a terminal window (under Applications-Utilities).

ssh your-username@pod.cnsi.ucsb.edu   or the other clusters, eg. ssh your-username@braid2.cnsi.ucsb.edu

Also, note you will need to be on campus, or, if you're off campus, you'll need to use the VPN

Running jobs

We use Slurm on the clusters to submit and run jobs.  Basically you need to create a simple script file to tell the compute node what to do.  A simple example is here (assuming the code you want to execute is named 'a.out' and you want to load the Intel 19 compilers via a module)

#!/bin/bash 
#SBATCH -N 1 --cores=1 
cd $SLURM_SUBMIT_DIR 
module load intel/19 
./a.out 

there's quite a few Slurm variables you can set - see this expanded example

#!/bin/bash
#SBATCH -N 1 --cores=1
#SBATCH --time=01:00:00   #  Set the CPU time limit - here, as an hour
#SBATCH --job-name="testslurm"
#SBATCH --mail-type=begin    # Send me an email when the job starts
#SBATCH --mail-user=myemailaddress@ucsb.edu  # where to email!

### next few only useful on Braid2
#SBATCH --mem=1500G             #  How much RAM - if you need the 3TB node, instead of one (and using 'largemem' partition)

#SBATCH --constraint="Intel"    #  Can constrain to Intel or AMD types of CPUs.  Only use if necessary (braid2/largemem).

cd $SLURM_SUBMIT_DIR

module load intel/19

./a.out

A few other special Slurm settings:

For GPU jobs, you need to 'reserve' a GPU.

#SBATCH --gres=gpu:a40:4  (on braid2...on pod, just --gres=gpu:1 - they are all the same type of GPUs)

Note that you can request memory, say you want the higher memory 'batch' nodes (on braid2 - on Pod, all nodes have same, 192GB, of RAM)

#SBATCH --mem=400G