Introduction to Kellogg Linux Cluster

Introduction to Kellogg Linux Cluster#

The Kellogg Linux Cluster (KLC) is a set of high-memory Linux servers available to Kellogg researchers for interactive computing, data analysis, and running computationally intensive jobs. If your analysis is too slow on a laptop, requires more memory than you have locally, or needs to run overnight, KLC is the right tool.

What KLC Offers#

Large memory — each node has 1.5–2 TB of RAM, far more than any laptop
Many CPU cores — up to 64 cores per node for parallel workloads
Large storage — 2 TB of project storage per researcher, separate from your home directory
Shared datasets — curated research datasets pre-loaded and ready to use
Scientific software — the same software library as Northwestern Quest, available via module load
Reproducible workflows — run the same code on the same hardware every time

KLC Architecture#

KLC consists of 12 high-memory Linux nodes:

Nodes	CPU Cores	RAM
klc0304 – klc0307, klc0401, klc0402, klc0503 (latest gen)	64 cores	2 TB
klc0202, klc0203, klc0301 – klc0303	52 cores	1.5 TB

All nodes share the same file systems, so a file saved on one node is accessible from any other.

Usage policy: Each user may use up to 24 CPU cores concurrently at normal priority. Going beyond this reduces priority for all your processes. Contact rs@kellogg.northwestern.edu if your work regularly needs more than 24 cores.

KLC Reserve#

KLC Reserve gives you access to dedicated GPU and high-core-count CPU nodes through the SLURM job scheduler — resources not available on the standard KLC login nodes.

Resource Type	Best For
GPU — H100	Large-scale LLM training and inference, deep learning
GPU — A100	GPU-accelerated computation, ML training
GPU — L40S	LLM inference, rendering, general GPU workloads
High-memory CPU	Very large in-memory datasets, parallel batch jobs

Jobs are submitted to the kellogg SLURM partition and run exclusively on allocated resources for the duration of your job. No additional account setup is needed if you already have KLC access. Full details: KLC Reserve

Getting Started: Step by Step#

Step 1 — Get an Account#

If you do not already have a KLC account, contact Kellogg Research Support at rs@kellogg.northwestern.edu .

Step 2 — Connect to KLC#

KLC OnDemand (zero setup — start here if you’re not sure)

Go to Quest OnDemand , log in with your NetID, and select the Kellogg Linux Cluster profile. From there you can launch Jupyter, RStudio, VS Code, or a full graphical desktop directly in your browser — nothing to install or configure. Full walkthrough: KLC OnDemand

SSH (command line)

On Mac, open the built-in Terminal app and run:

ssh your-netid@klc0305.quest.northwestern.edu

On Windows, open PowerShell and run:

ssh your-netid@klc0305.quest.northwestern.edu

Set up passwordless SSH login to avoid entering your password each time. Full instructions: SSH

VS Code with Remote SSH (recommended for development)

Connect VS Code directly to KLC and edit files, run notebooks, and use GitHub Copilot — all from your laptop. See VS Code Workflow for KLC.

FastX (full graphical desktop)

For software that requires a full X11 desktop — MATLAB, Stata, SAS. See FastX.

Step 3 — Understand Your Storage#

Once connected, you have two storage locations:

Location	Path	Quota	Purpose
Home directory	`/home/your-netid/`	80 GB	Personal files, config, small scripts
Project directory	`/kellogg/proj/your-netid/`	2 TB	Data, conda environments, job output

Always store data and environments in your project directory — the home directory fills up quickly.

Full details: KLC Filesystem

Step 4 — Set Up a Python or R Environment#

KLC uses mamba (a faster version of conda) to manage software environments. Create an isolated environment in your project directory:

module load mamba/24.3.0
mamba create -p /kellogg/proj/your-netid/envs/my-project python=3.11
source activate /kellogg/proj/your-netid/envs/my-project
mamba install pandas numpy matplotlib

Full details: Conda Environments on KLC

Step 5 — Run Your First Script#

With your environment active, run a Python script:

python my_analysis.py

For long-running jobs, wrap your session in tmux and save a log file so the job keeps running if you disconnect:

tmux new -s my-job
mkdir -p logs
python my_analysis.py 2>&1 | tee -a logs/my_analysis.log
# Press Ctrl+B then D to detach — job keeps running
# Reconnect later with: tmux attach -t my-job

Full details: Launching Jobs · Using tmux

Key Things to Know#

Stay within the 24-core limit per user across all KLC nodes. For jobs that should survive a disconnect, use tmux and save output to a log file — see Launching Jobs.
Use your project directory for data and environments, not your home directory.
Load software with module load before using R or Stata: module load R/4.5.1, module load stata/17. For Python, load mamba/24.3.0 and activate a conda environment — see Step 4 above.
Check memory usage before starting large jobs: module load glances && glances shows all users’ CPU and memory in real time.

Where to Go Next#

Goal	Page
Connect and log in	Access KLC
Set up a development environment	Conda Environments
Transfer files to/from KLC	Transferring Files
Keep jobs running after disconnect	Using tmux
Full reference documentation	KLC User Guide
Use VS Code with KLC	VS Code Workflow
Use LLMs on KLC	Open Source LLMs
Interactive vs. scheduled jobs	When to Use KLC Reserve
Submit SLURM and GPU jobs	KLC Reserve
CPU vs. GPU concepts	GPU Concepts and Options
SLURM job scheduler (deep reference)	Quest SLURM docs
Software modules reference	Quest Modules docs
Filesystem quotas and permissions	Quest Filesystem docs
Login methods (SSH, FastX, OnDemand)	Quest Login docs