|
"Research computing" involves running specialized software and operating systems to analyze complex datasets. Because these systems often utilize command-line terminals rather than typical consumer interfaces, they can be intimidating to beginners. This page outlines the shared computing infrastructure, applications, and onboarding support structures available to the Geneseo community.
|
Table of Contents
|
If you are new to working within terminal environments or remote servers, these training materials can help you get oriented:
Through NSF grant programs that mandate cross-institutional resource sharing, Geneseo users have zero-cost access to high-performance federal cloud systems. These utilize the InCommon single sign-on integration. Please contact CIT to request provisioning onto:
- ACCESS: Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support ecosystem.
- National Research Platform (NRP): Distributed cluster offering interactive app spaces (like Jupyter Notebooks and RStudio) backed by remote graphics processing hardware.
- Chameleon: A deeply reconfigurable bare-metal cloud platform architecture built for large-scale computer science testing.
CIT manages two shared, highly-optimized hardware nodes explicitly configured for processing massive computational workloads and deep learning modeling sequences.
Network & Authorization Pre-requisites
- Network Isolation: You must be physically connected to the campus network environment or actively tunneled via the Geneseo VPN client to establish a connection.
- Access Management: Identity permissions are bound exclusively to the
Lambda-User Active Directory security group. Please submit a request to Ask CIT for group enrollment.
Workstation 1: deeplearning1 (Linux)
- Hostname:
deeplearning1.geneseo.edu | Network Address: 137.238.8.47
- Platform Environment: Ubuntu Linux 20.04 LTS
Method A: Experimental Web Interface
Note: This interactive web environment does not support the Apple Safari web browser.
- Navigate to Xpra HTML5 Client
- Ensure the target settings are mapped to — Host:
deeplearning1.geneseo.edu | Port: 443.
- Provide your standard Geneseo short username and network password.
- Click Connect to automatically construct and route into your private container process.
Operational Guidelines:
- App Initialization: Use the top-left 3-line hamburger icon. Standard applications are located under Start > Geneseo.
- Persisting Workflows: To leave code execution active in the background, exit your browser tab or select Hamburger > Disconnect.
- Termination: To completely teardown your active session instance, choose Hamburger > Server > Shutdown Server.
- Data Transfer: Drag and drop local assets onto the viewport to stream them into your user account's
Downloads partition. To push down a server file to your local drive via the web portal, execute: xdg-open /path/to/remote/file inside the active console.
Method B: Native SSH Interface
The Linux environment hosts an active SSH server mapped to TCP port 22. Confirm one of these fingerprint identities upon primary authentication:
SHA256:laNR7b1nLDAkCxFF0JMpzGyTCsPPQbjj+TOb9HuSfNQ (RSA)SHA256:uVuGpGGB9U5+OEBpKLfatIkUoCf09TjxycK+Ksg6evg (ECDSA)SHA256:1C8/YeaK61hFXIfjzjUW4hlJBche1nGs/C19/4U081g (ED25519)
X11 Graphical Forwarding over SSH:
- Windows: Modern Windows 10/11 iterations natively forward remote X11 elements if utilizing the Windows Subsystem for Linux 2 (WSL2) platform layer.
- macOS: Requires downloading and running XQuartz to pipe application interfaces downstream to your display engine.
- Syntax: Append the X11 allocation flag during execution:
ssh your-username@deeplearning1.geneseo.edu -Y. Test validation by calling xlogo.
Workstation 2: deeplearning2 (Windows)
- Hostname:
deeplearning2.geneseo.edu | Network Address: 137.238.8.48
- Platform Environment: Windows 10
- Access Vectors: Access uses Remote Desktop Protocol (RDP) hitting TCP port 3389. Mac platforms should utilize the official Microsoft Remote Desktop companion app.
- Concurrency Limitations: The host OS platform strictly limits terminal sessions to **one single concurrent operator**. You must completely Log Out of the system when exiting your workspace. Closing the client window without logging out keeps your session active, completely blocking out all other researchers.
The deeplearning1 Linux platform environment comes loaded with analytical processing foundations:
- Languages & Runtimes: Python 3.8, R 4.3.x, Mathematica 13, and Matlab R2023a (accessible inside
/usr/local/MATLAB).
- Lambda Stack Bundle: Core machine learning environments backed by hardware-accelerated drivers (TensorFlow, Keras, PyTorch, Caffe, Caffe 2) driven via optimized **CUDA** and **cuDNN** binaries.
- Specialty Engines: 3DSlicer 5, FSL, and the astronomical Image Reduction and Analysis Facility (IRAF).
Miniconda Architecture
To declare and isolate distinct runtime dependency trees without breaking core system paths, utilize Miniconda:
1. Inject the environmental path into your active terminal shell execution loop:
source "/deeplearning/miniconda3/etc/profile.d/conda.sh"
2. Enumerate existing environments:
conda env list
3. Mount a pre-existing target runtime workspace:
conda activate [target-environment-name]
4. Build an isolated custom python workspace framework (e.g., Python version 3.9):
conda create -n [your-env-name] python=3.9
Both processing nodes leverage matching robust high-compute server configurations:
- Compute: 12-core Intel Xeon CPU E5-2650 v4 clocked @ 2.20GHz tied to 128 GB of RAM.
- Acceleration Array: 4x discrete NVIDIA Titan V GPUs (delivering 12 GB VRAM and 3,584 CUDA parallel processing cores per card).
- Storage Elements: 1TB solid-state OS boot architecture, combined with a high-throughput scratch array consisting of 3x 4TB enterprise SSD units orchestrated under a ZFS Raidz1 tier, available at
/deeplearning.
Still Need Help?
Ask CIT! 📞 (585) 245-5588 | 📧 Email | 💬 Chat | 📝 Submit a Request and we'll be happy to assist you.