Skip to content

🖥️ cloudworkstation

Interactive research workstations in the cloud

Visit cloudworkstation.io → | GitHub →

Overview

cloudworkstation provides pre-configured cloud workstations for data science, machine learning, and computational research. Launch GPU-powered environments in minutes with common tools, frameworks, and libraries pre-installed.

Key Features

Pre-Configured Environments

  • Machine Learning - PyTorch, TensorFlow, JAX, scikit-learn
  • Data Science - pandas, NumPy, Matplotlib, Seaborn
  • Deep Learning - CUDA, cuDNN, GPU-optimized frameworks
  • Development Tools - VSCode, Jupyter, terminals

GPU Support

  • NVIDIA GPUs - T4, V100, A10G, A100 instances
  • CUDA Ready - Pre-installed CUDA toolkit and drivers
  • Multi-GPU - Support for multi-GPU training
  • Cost Options - On-demand or spot instances

Workspace Management

  • Quick Launch - Start workstations in under 2 minutes
  • Persistent Storage - EBS volumes for data persistence
  • Snapshots - Save workspace state for later resumption
  • Auto-Shutdown - Automatic shutdown to control costs

Collaboration

  • Shared Workspaces - Collaborate with team members
  • Access Control - IAM-based permissions
  • Session Sharing - Share terminal sessions
  • Data Sharing - Shared S3 buckets for datasets

Use Cases

Interactive Machine Learning

Develop and train ML models with GPU acceleration without managing infrastructure.

Data Exploration

Analyze large datasets interactively with pre-installed data science tools.

Model Prototyping

Quickly prototype deep learning models with access to latest frameworks.

Teaching and Workshops

Provide students with consistent, pre-configured environments for courses.

Architecture

graph TB
    User[Researcher] -->|Launch| CW[cloudworkspaces]
    CW -->|Create| EC2[GPU EC2 Instance]
    EC2 -->|Mount| EBS[Persistent EBS Volume]
    EC2 -->|Access| S3[S3 Data Buckets]

    EC2 -->|VSCode| Browser[Web Browser]
    EC2 -->|Jupyter| Browser
    EC2 -->|SSH| Terminal[Local Terminal]

    style CW fill:#fff4e1
    style EC2 fill:#e1f5ff

Available Workstation Types

ML Development (ml.*)

  • ml.g4dn.xlarge - 1x NVIDIA T4, 4 vCPU, 16 GB RAM
  • ml.g5.xlarge - 1x NVIDIA A10G, 4 vCPU, 16 GB RAM
  • ml.p3.2xlarge - 1x NVIDIA V100, 8 vCPU, 61 GB RAM

Deep Learning Training (dl.*)

  • dl.g5.2xlarge - 1x A10G, 8 vCPU, 32 GB RAM
  • dl.p3.8xlarge - 4x V100, 32 vCPU, 244 GB RAM
  • dl.p4d.24xlarge - 8x A100, 96 vCPU, 1152 GB RAM

Data Science (ds.*)

  • ds.m5.2xlarge - 8 vCPU, 32 GB RAM (CPU-only)
  • ds.r5.4xlarge - 16 vCPU, 128 GB RAM (memory-optimized)

Development (dev.*)

  • dev.t3.large - 2 vCPU, 8 GB RAM (cost-optimized)

Getting Started

Launch a Workstation

# Install cloudworkstation CLI
brew install cloudworkstation

# List available workstation types
cloudworkstation list-types

# Launch ML workstation
cloudworkstation launch \
  --type ml.g4dn.xlarge \
  --name my-ml-workspace \
  --storage 100

# Connect via VSCode
cloudworkstation connect my-ml-workspace --vscode

# Or via SSH
cloudworkstation ssh my-ml-workspace

Manage Workstations

# List running workstations
cloudworkstation list

# Stop workstation (preserves data)
cloudworkstation stop my-ml-workspace

# Start stopped workstation
cloudworkstation start my-ml-workspace

# Create snapshot
cloudworkstation snapshot my-ml-workspace --name experiment-1

# Delete workstation
cloudworkstation delete my-ml-workspace

Pre-Installed Software

Python Ecosystem

  • Python 3.11, 3.10, 3.9 (via conda)
  • PyTorch 2.1 (GPU)
  • TensorFlow 2.14 (GPU)
  • JAX (GPU)
  • scikit-learn, XGBoost, LightGBM

Development Tools

  • VSCode Server
  • Jupyter Lab
  • Git, GitHub CLI
  • Docker

Data Tools

  • pandas, Polars
  • DuckDB
  • Apache Arrow
  • Parquet tools

Utilities

  • tmux, screen
  • htop, nvtop
  • AWS CLI
  • S3 mount utilities

Integration with ResearchComputing

Data Sources

  • S3 - Direct access to S3 buckets for datasets
  • cargoship - Retrieve archived data for analysis

Workflow

  1. Launch cloudworkstation for interactive development
  2. Prototype and test on small datasets
  3. Scale to atom for production HPC runs
  4. Archive results with cargoship

Account Management

  • Provision accounts via petri
  • Cost tracking and budget alerts

Documentation

Technology Stack

  • Language: Go (CLI), Python (utilities)
  • Infrastructure: EC2, EBS, CloudFormation
  • Networking: VPC, Security Groups
  • Storage: EBS, S3
  • Access: Systems Manager Session Manager, SSH

Project Status

Current Version: v1.2.0 Status: Production Ready License: Apache 2.0

Contributing

Contributions welcome! See the contribution guide.

Support

← Back to Ecosystem