HPC and Infrastructure

1 Why HPC is the lab default

Harvard University’s computing cluster is called FASRC (Faculty of Arts and Sciences Research Computing), and is one of the largest, most powerful compute clusters on the East Coast. It provides scalable computing resources, secure environments for sensitive data, and a shared infrastructure that supports collaboration and reproducibility. As discussed in Principle 1, we treat HPC as the default environment for data science in the lab. If your project works on HPC, it is more likely to be reproducible, shareable, and scalable. If it only works on your local machine, which only has settings and configurations specific to you, it is more likely to break when you or someone else tries to run it again in the future. By working this way, we intentionally introduce a small amount of friction upfront, to save a lot of time and headache in the future.

But don’t let that intimidate you! The FASRC team has done a great job of making the cluster accessible and user-friendly, and there are many ways to interact with it that require close to zero setup or expertise. The goal of this guide is to help you find the right access path for your needs, and to give you some practical tips and examples for working on HPC.

2 Some Lingo

Let’s clear the air of some jargon that you might encounter when working with HPC:

Term Definition
HPC High Performance Computing. A general term for powerful computing resources that can handle large-scale data analysis and modeling.
FASRC Faculty of Arts and Sciences Research Computing. The computing cluster used at Harvard.
Cannon The standard FASRC cluster for general research computing. Synonymous with FASRC for all intents and purposes.
FASSE The secure environment for work that requires stronger data protection (L3 and above).
Open OnDemand (OOD) A web-based interface that provides access to interactive apps, file browsing, and job submission tools on FASRC.
Shell/SSH (noun) A command-line interface used to interact with the operating system and run commands, including managing files and submitting jobs. Common shells include bash (default for FASRC) and zsh (default for macOS).
SSH (verb) To connect to a remote machine using the SSH protocol (e.g., “SSH into the cluster”).
Slurm A workload manager that allows you to schedule and manage batch jobs on the cluster.
Batch Job A task given to the system in the form of a script. You submit it to the queue and it runs when resources are available. For example, a Python script that you want to run on the cluster without manual intervention.
login node The part of the cluster where you can run commands, manage files, and submit jobs. It is not meant for heavy computation.
compute node The part of the cluster where more intense computation happens. Usually, you use this to do long-running interactive tasks like RStudio or Jupyter notebooks, and submit batch jobs.

🚦🚧 This section is under construction 🚧🚦

Another confusing collection of jargon is the interchangeable use of environments, software, and modules. Let’s clarify these:

  • Software: The actual programs and tools that you use to do your work, such as R, Python, or specific packages. For example, R version 4.2.0 is a piece of software.

  • Modules: A system for managing software on HPC clusters. Modules allow you to load and unload specific versions of software and their dependencies. For example, you might run module load R/4.2.0 to load R version 4.2.0 and its associated libraries into your environment.

  • Computing Environment: The overall setup of your computing context, including the hardware, operating system, software packages, and modules that are available to you. For example, the Cannon cluster is an environment, and the FASSE cluster is a different environment with different security settings and software availability.

  • Environment variables: These are key-value pairs that are set in your shell and can be accessed by your programs. They often contain configuration settings, such as paths to software or proxy settings. Importantly, environment variables are “set” by the shell when it starts, are specific to the user who starts the shell, and are inherited by any processes that are started from that shell. They are not global settings for the cluster, but rather specific to your session. For example, when I log into the cluster, my shell automatically sets environment variables that point to the R and Python installations, and I can also set my own environment variables for things like proxy settings or custom paths. These settings are made in the file ~/.bashrc, which you can modify to create a customized environment. Take a moment to look at your own environment variables by running env in your shell, to see what is set by default, and cat ~/.bashrc to see your customizations.

Tip

For your purposes, “Software” refers broadly to the programs and tools you use, “Modules” are the FASRC-managed software packages you load, “Computing Environment” is the overall context of your work on HPC, and “project environment,” is the specific configuration of software and settings that your local project relies on to run successfully.

3 Interface

We recommend two main interfaces for working on FASRC: Open OnDemand (OOD) and VSCode1. OOD is a web-based interface that provides access to interactive apps like RStudio Server, Jupyter, and Matlab, as well as file browsing and job submission tools. VSCode is a powerful code editor that can connect to the cluster via SSH, allowing you to edit files, run commands, and manage your code directly from your local machine. Both interfaces have their strengths and weaknesses, and the best choice depends on your specific needs and preferences.

4 Choose the right access path

content

Ultimately, this is just submitting a job that you interact with through the browser.

content

vscode remote ssh is complicated

🚦🚧 This section is under construction 🚧🚦

Footnotes

  1. There is technically a third interface, which is the command-line shell accessed via SSH. However, we recommend using OOD or VSCode for most interactive work, as they provide a more user-friendly experience and better support for code editing and project management. Doing so purely via SSH is possible, but requires strong familiarity with the command-line, and doesn’t really provide any advantages over the other two options for most use cases.↩︎