Contents

• 1. NVIDIA Magnum IO GPUDirect Storage Overview Guide
• 2. Introduction
  • 2.1. Related Documents
  • 2.2. Benefits for a Developer
  • 2.3. Intended Uses
• 3. Functional Overview
  • 3.1. Explicit and Direct
  • 3.2. Performance Optimizations
    • 3.2.1. Implementation Performance Enhancements
    • 3.2.2. Concurrency Across Threads
    • 3.2.3. Asynchrony
    • 3.2.4. Batching
    • 3.2.5. Use of CUDA Streams in cuFile
  • 3.3. Compatibility and Generality
  • 3.4. Monitoring
  • 3.5. Scope of the Solutions in GDS
  • 3.6. Dynamic Routing
    • 3.6.1. cuFile Configuration for Dynamic Routing
    • 3.6.2. cuFile Configuration for DFS Mount
    • 3.6.3. cuFile Configuration Validation for Dynamic Routing
• 4. Software Architecture
  • 4.1. Software Components
  • 4.2. Primary Components
    • 4.2.1. Workflows for GDS Functionality
    • 4.2.2. Workflow 1
    • 4.2.3. Workflow 2
  • 4.3. Aligning with Other Linux Initiatives
• 5. Deployment
  • 5.1. Software Components for Deployment
  • 5.2. Using GPUDirect Storage in Containers
• 6. Notice
• 7. OpenCL
• 8. Trademarks