DRA
Before
After
You ask for: nvidia.com/gpu: 1
There is no way to say:
- I need this type of device
- with these capabilities
- prepared in this way
Instead of requesting a number you describe what you actually need...
You can ask for: nvidia.com/gpu: 1, product ID A100-SXM4-40GB of memory and 3456 FB64 cores
- ResourceClaim = what you want
- Driver = what exists
- Scheduler = matches them
resources:
limits:
nvidia.com/gpu: 1spec:
requirements:
- deviceClassName: gpu
selectors:
- name: model
value: A100
- name: memory
value: "40Gi" +-------------------+ +---------------------+ +----------------------+
| Pod | | ResourceClaim | | DRA Driver |
|-------------------| |---------------------| |----------------------|
| needs: | -----> | requests: | -----> | knows real devices |
| - "GPU workload" | | - device type | | - GPU A100 |
| | | - capabilities | | - FPGA X |
| | | - constraints | | - topology info |
+-------------------+ +---------------------+ +----------------------+
| | |
| v |
| +---------------------+ |
| | Scheduler | |
| |---------------------| |
| | finds matching node | |
| +---------------------+ |
| | |
v v v
+--------------------------------------------------+
| Allocated & Prepared Device |
| (correct hardware, configured, ready to use) |
+--------------------------------------------------+
↓
🚀 Pod starts
DRA
By fmuyassarov
DRA
- 8
