M3 High-Performance Computing (HPC) Quickstart
Last update: 15 June 2017
In this tutorial, we're going to practice doing a BEAST run on the HPC. The BEAST software computes evolutionary trees from input sequence data, but no knowledge of the BEAST software is required in this tutorial. We're going to: - Prepare a SLURM script,
my_slurm_script.txt - a bunch of instructions that tells the HPC what to do - Get an
benchmark1.xml - a bunch of instructions that tells the BEAST software what to do. Download it here - this file contains the raw data and some parameters in a format that BEAST understands. - Upload both files to the HPC.
Cyberduck is a point-and-click interface which helps you do this; get it here - Submit the SLURM script to M3, and review a bunch of frequently used HPC commands on the terminal - Retrieve your job output after the run is complete.
A generic programming script is a text file containing a list of instructions for a computer to carry out. The computer must be given instructions in a very specific format that it can understand.
M3 is the high-performance computing (HPC) resource provided by Monash. It's like having a second very, very powerful computer (actually computers), but you can't easily interact with it with a point-and-click interface. You have to tell it what to do by preparing a list of instructions, called a SLURM script. Note that "HPC" and "M3" will be used interchangeably where context is clear, where "M3" is actually the name of Monash Uni HPC resource.
SLURM is a queueing manager used by many HPCs around the world; widely used because it's open-source (i.e. free). It's necessary because there are often multiple users requesting computational resources from the same HPC, with different computational requirements, runtimes, and so on. So SLURM doesn't actually do any computation, it assigns user-submitted computing tasks to the nodes within the HPC cluster as efficiently as possible.
2. Uploading Your Files to M3
To log in to M3 through Cyberduck, click the
Open connection button at the top left of the window, and fill in the fields as follows:
(Of course, for the
password fields, use your own username and password.
Uploading your files is then just a simple matter of drag-and-drop.
Alternatively, you can upload files on the terminal using
scp (secure copy) like so:
scp path\to\file\on\your\machine email@example.com:folder\on\massive
3. Writing a SLURM Script
M3 needs two kinds of information: 1. Specifics of your HPC computational needs like number of nodes, number of cores per node, memory usage, runtime etc. For the purposes of this tutorial, you can just use the settings provided in the example below. 2. The actual job that needs to be run, e.g. a command like
beast -beagle_off my_input_file.xml
The example SLURM script below will do a BEAST run using the
benchmark1.xml file downloaded earlier. On your own computer, create a new SLURM script by copying and pasting the text below into a new text file, and save that text file as
my_slurm_script. Make sure to enter your email at the line
#!/bin/bash #SBATCH --job-name=my_job #SBATCH --ntasks=1 #SBATCH --ntasks-per-node=1 #SBATCH --time=0-02:00:00 #SBATCH --mail-user=<your_email> #SBATCH --mail-type=END #SBATCH --mail-type=FAIL module load beast1/1.8.4 module load beagle beast -beagle_off benchmark1.xml
To break that down: - The very first line
#!/bin/bash tells the HPC that this file is a list of instructions, not just a bunch of text. Always include it in the first line of your SLURM scripts. - All the lines starting with
#SBATCH are run parameters, e.g. run time, number of CPUs you'd like to request, etc. - The lines at the bottom:
beast... are the actual BEAST command lines. That is, if you have BEAST on your own computer, you'd run this same computation using
beast -beagle_off benchmark1.xml on your own command line.
Further details on what all that means will be covered in another tutorial.
4. Logging into to M3
Open your Terminal (assuming you're using a Mac), and in the command line, login using:
$ ssh <your_username>@m3.massive.org.au
You'll be prompted for your password, and then you're in. If you type
-ls to see what you have, you should see your own folder, plus a bunch of files that're basic SLURM scripts, provided by the folks at M3. Your M3 directory is, for the moment, just one folder.
benchmark1.xml to your account on M3, using Cyberduck. Now submit your script to the queue by entering, in the command line:
$ sbatch my_slurm_script
You should recieve a message containing a job ID, like:
Submitted job 1234
It usually takes a second to submit your script to the queue.
Note that if you don't see your script in the queue, two things could have happened: 1. your script still hasn't been submitted yet, or 2. your script terminated almost instantaneously with an error, and therefore appeared in and then disappeared from the queue within microseconds. To check the second scenario, check if there's a slurm output file in your directory. You can also tell SLURM to send you an email whenever the job ends, whether it ended successfully, or with an error.
5. Reviewing your Output
All jobs that were run on the HPC will automatically generate an output file
slurm-<job_id>.out. This is like a text file, but, somewhat annoyingly, can't be viewed with a text editor. You can read the contents of this file using the
BEAST would have generated some other files as well, but we don't need to look at those for the purposes of this tutorial. The main BEAST output would have been recorded in the SLURM output file.
Upon completion, you should recieve an email from M3, because you asked for one in the SLURM script above.