GPU Cluster is an internal web-based system used to manage and automate access requests to GPU clusters operated by the Computer Science Faculty, Technion.
The system provides a controlled workflow for requesting, approving, and managing GPU cluster access for students, researchers, and staff across multiple faculties.
Purpose
- Centralize and automate GPU cluster access requests
- Enforce eligibility, role, and supervisor-based policies
- Provide transparent request lifecycle management
- Reduce manual handling and duplicate requests
Key Capabilities
- Authenticated access using Technion accounts
- Role-based workflow (Admin / Supervisor / User)
- Cluster-aware requests (e.g. Newton, Darwin, Lambda, GipDeep)
- Supervisor approval model with validation
- Request lifecycle management (Pending, Active, Inactive, Rejected)
- Controlled reuse of rejected requests to avoid duplication
- Email notifications for request status changes and login events
- Export of request data (Excel) for administrative review
Access Logic (High-Level)
- Users authenticate with their Technion credentials
- Available clusters and request options are determined by:
- User affiliation (faculty / role)
- Academic status (undergraduate / graduate / staff)
- Supervisor eligibility
- Requests are reviewed and managed through a structured approval workflow
Design Principles
- Policy-driven access control
- Clear separation between user, supervisor, and administrator actions
- Emphasis on transparency and traceability
- Production use with continuous incremental improvements
Status
GPU Cluster is actively used in the production HPC environment of the Computer Science Faculty and continues to evolve based on operational feedback.
Tool issue or incorrect output?
If this tool behaves incorrectly or shows inconsistent data, please report it to the DevOps team: devops@cs.technion.ac.il