5 Things You Need to Know about the Computing Cluster

This is a practical tutorial on how to use Sherlock in your research. Sherlock is a computing cluster open to researchers at Stanford University. It's based on the slurm scheduler environment that handles all the computing resources for you. The only thing we need to do is to submit computing jobs to the cluster and the rest will be handled by slurm. You can submit hundrets or even thousands of jobs to the cluster. This means that you can compute things on the cluster in hours that would take weeks to compute on a laptop.

1. Request Acccount

To get access to the cluster, you need to request an account here.

2. Upload Your Script

The best way to upload your script to Sherlock is through GitHub. To start install the GitHub desktop client on your laptop. Then clone this project with File -> Clone Repository.... Now open the file called submit.sbatch with your favorite text editor, e.g Atom is a good one. This code is now on your laptop, but how can you get it onto the cluster? We can use GitHub for that, but first we need to login to Sherlock. Open a Terminal windows and type:

ssh login.sherlock.stanford.edu

Now go to $SCRATCH folder on Sherlock. This is where you have unlimited storage space. So this will be good for storing all your fcs files. You still need to backup your fcs somewhere because $SCRATCH is only meant for temporary storage of files.

cd $SCRATCH

To check where you are, you can use the command pwd to print out the current directory. Now we can download code from GitHub into the $SCRATCH folder.

git clone https://github.com/ChristofSeiler/sherlock.git
cd sherlock

Pro tip: If you don't want to retype your password everytime, you can use a credential helper.

3. Upload Your Data

To upload your own data, e.g. sample info tables or fcs files. You can use the following the scp command. Open a new terminal go to the folder where you local file is located and type:

scp you_local_file /scratch/users/your_username/.

In this tutorial, we don't have any data to uplaod.

4. Run Your Script

If your script knits an Rmd then we also need to install pandoc as described here.

Ok, now we are ready to submit our first job to the cluster. All the information is in the file submit.sbatch. It will ask the cluster for a certain amount of memory, time, and number of processors. Then it will get into the scheduler queue and we will have to wait until it's our turn. Depending on how much computing resources we need this will take more or less time. But usually on the new Sherlock 2.0 cluster it's pretty fast.

Command to submit a job:

sbatch submit.sbatch your_rmd_filename

You can also get an email notification from the cluster when the job is done or failed like this:

sbatch --mail-user=your_email_address submit.sbatch your_rmd_filename

To check the status of your job:

squeue -u your_username

When the job is completed you will see the result in a text file with extension .out. To open it type:

more slurm-*.out

5. Download Your Results

You can download your result to your laptop by switching to you other terminal and typing:

scp login.sherlock.stanford.edu:/scratch/users/your_username/sherlock/slurm-*.out .

Cache Login Credentials

To avoid retyping your login thousand times do this on your laptop:

mkdir -p ~/.ssh
echo "Host sherlock sherlock.stanford.edu sherlock* sherlock*.stanford.edu
  GSSAPIDelegateCredentials yes
  GSSAPIAuthentication yes
Host login.sherlock.stanford.edu
  ControlMaster auto
  ControlPersist yes
  ControlPath ~/.ssh/%l%r@%h:%p" >> ~/.ssh/config

Using a Graphical User Interface

For steps 2, 3, and 5 you can also use a webinterace called fetch or Globus.

• Instructions for fetch: here
• Instructions for Globus: here

When transfering files using Globus enable encrypt transfer in Transfer Settings.

Everything within RStudio

See instructions here.

Advanced Stuff

If you want to use the cluster wihtin an R script read this.