Skip to content

HPCsetup.md

Latest commit

 

History

History
75 lines (51 loc) · 3.25 KB

HPCsetup.md

File metadata and controls

75 lines (51 loc) · 3.25 KB
Raw

HPC Setup

These instructions were modified from Jesse Cusack's scientific-workflow.

Quick access to the server

To more quickly access the server, you can set up ssh keys on your local machine, add login credentials to your config file, and create profiles in iTerm2 with different colors and shortcuts.

Working with GitHub on the server

To clone your project repository on the server, you will need to create an ssh-key on the server, following the instruction on GitHub. Make sure to select the Linux tab.

Installing Miniconda on the server

Log into the HPC cluster. To download Miniconda, run:

cd ~
wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

To install Miniconda, run:

bash Miniconda3-latest-Linux-x86_64.sh

And follow the prompts to install it in your home directory (this usually just means following the default options).

As part of the installation, you may be asked for permission to modify your terminal configuration (the .bashrc file that lives in your home directory). This is recommended, so that every time you log into the HPC, you will have access to the conda commands automatically.

Upon successful installation, I would recommend setting conda-forge as the default channel package searches. I find that it has a more complete list of packages than the standard conda channel. To do so, run the following in the terminal:

conda config --add channels conda-forge
conda config --set channel_priority strict

If the conda command is not recognized when you try to use it, you might have to add the path to ~/miniconda3/bin/ to your .bashrc file.

  1. Open ~/.bashrc with your preferred editor. Note that the example uses vi.
vi ~/.bashrc
  1. Add the path to your miniconda3/bin folder to your .bashrc by including the following line (replace /path/to/miniconda3/bin with the actual path, perhaps ~/miniconda3/bin):
export PATH="/path/to/miniconda3/bin:$PATH"
  1. Source the configuration or restart your Terminal session:
source ~/.bashrc

Working with Jupyter Lab

[Add instructions after testing them.]

CEOAS partitions

CEOAS has numerous partitions. Which one you should use depends on your work:

  • ceoas : general x86 CPU-only servers

  • ceoas-arm : ARM servers (namely the Grace Hoppers)

  • ceoas-gpu : x86 servers with GPUs

  • brizo : machines owned in part by LOL

  • PI machine partitions (brizo, ewg, etc.)

  • ceoas-lowprio, ceoas-arm-lowprio, ceoas-gpu-lowprio : PI machines at lower priority (i.e., will run if no higher priority jobs are submitted)

  • ceoas-interact : primarily for testing; jobs are limited (no exclusive, maximum of 8 cores and 64GB of RAM)

Nodes that we like to use

  • brizo : [add specs], new nodes, quite large, typically reserved for ocean simulations and not particle tracking
  • amaterasu : 2x8-core processors
  • yatagarasu : 2x12-core processors