FLARE (Federated Liquid Resources Exchange) is a GPU pooling platform for AI and HPC applications. Built on FLUIDOS, it enables dynamic GPU sharing across cloud providers through intent-based allocation and federated multi-tenant cloud architecture.
⚠️ Research Project
This is a research project and is not intended for production use. This project explores concepts in federated GPU resource management and serves as a proof-of-concept for distributed computational resource sharing. If you are interested in using these concepts for your commercial project or want to borrow ideas from this research, please contact CLASTIX.
- Executive Summary
- Problem Statement and Solution
- FLARE Architecture on FLUIDOS
- Technical Differentiation
- Implementation Roadmap
- Additional Resources
- Overview
- Prerequisites
- Development Environment Setup
- Configuration Summary
- Troubleshooting
- Cleanup
- Next Steps
- FLUIDOS Overview
- Prerequisites Check
- Provider Resources (Flavors)
- Consumer Resource Discovery
- Resource Reservation and Contract
- Establishing Cluster Peering
- Deploying Workloads on Virtual Nodes
- Multiple Virtual Nodes
- Multiple Virtual Nodes from Same Provider
- Troubleshooting Guide
- Quick Reference Commands
- Overview
- Prerequisites
- Development Timeline Context
- Workflow Evolution
- Workflow Comparison
- GPU Annotation Reference
- Use Cases and Examples
- Troubleshooting
- Cleanup
- Overview
- API Resources
- Workload Intent Schema
- Authentication
- API Endpoints
- Complete Examples
- Response Formats
- Error Codes
- Overview
- Quick Start
- Quick Reference Table
- Detailed Annotation Specifications
- Core GPU Annotations
- Location & Cost Annotations
- Performance Annotations
- GPU Sharing Annotations
- Network & Communication Annotations
- Provider Annotations
- Annotation Examples
- FLARE API Mapping
- Validation Rules
- Overview
- NVIDIA GPU Operator Label Reference
- Direct Label to Annotation Mappings
- Computed Annotation Mappings
- Complete Auto-Generation Example
- NVIDIA-Specific Extensions
- Node Selection
- Overview
- AMD GPU Operator Label Reference
- Direct Label to Annotation Mappings
- Computed Annotation Mappings
- Complete Auto-Generation Example
- AMD-Specific Extensions
- Node Selection
- Overview
- Prerequisites
- Hub Cluster Setup
- GPU Provider Setup
- Broker Requirements
- Verification and Testing
- Monitoring
- Troubleshooting
- AI Inference Service
- High-Performance AI Training
- LLM Fine-Tuning
- High-Performance Computing
- Real-Time Video Analytics
- Edge Inference
- Batch Processing
- Multi-Tenant Resources
- Distributed Workloads
- Cost Optimization Scenarios
- Executive Summary
- Project Timeline and Deliverables
- Results Achieved Against KPIs and Milestones
- Technical Innovations and Architecture
- Project Challenges and Technical Solutions
- Project Conclusion and Assessment
- Documentation License: Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)
- Source Code License: Apache License 2.0
- FLUIDOS Project - Base infrastructure platform
- Liqo Project - Multi-cluster connectivity layer
- Fake GPU Operator - GPU simulation for development and testing
- Capsule Project - Multi-tenancy solution for Kubernetes