PhoenixOS
0.1.0
Getting Started Guide for nVIDIA CUDA
Getting Started Guide for AMD ROCm
POS Architecture Overview
Checkpoint Mechanism
1. Checkpoint of Device Memory
2. Checkpoint of Driver Abstraction
3. Infer and Validation of Memory Access Pattern
4. Distribution Support
Migrate Mechanism
PhoenixOS
Checkpoint Mechanism
View page source
Checkpoint Mechanism
PhOS
adopts transparent yet non-stop-of-world mechanism to save GPU state.
1. Checkpoint of Device Memory
1.1. Checkpoint Stages
1.2. Checkpoint Pipeline
1.3. Checkpoint Deduplication
2. Checkpoint of Driver Abstraction
2.1. Checkpoint of Driver Abstraction
3. Infer and Validation of Memory Access Pattern
4. Distribution Support