| | 1 | = Work fetch and GPUs = |
| | 2 | |
| | 3 | == Current policy == |
| | 4 | |
| | 5 | * Do weighted round-robin simulation, get per-project and overall CPU shortfalls |
| | 6 | * If overall shortfall, get work from project with highest LTD |
| | 7 | * Scheduler request includes just "work_req_seconds". |
| | 8 | |
| | 9 | Problems: |
| | 10 | |
| | 11 | There may be no CPU shortfall, but GPU is idle |
| | 12 | |
| | 13 | If GPU is idle, we should get work from a project that potentially has jobs for it. |
| | 14 | |
| | 15 | If the project has both CPU and GPU jobs, we may need to tell to send only GPU jobs. |
| | 16 | |
| | 17 | LTD isn't meaningful with GPUs |
| | 18 | |
| | 19 | == New policy == |
| | 20 | |
| | 21 | Keep track of which projects can use which GPUs. |
| | 22 | |
| | 23 | Scheduler request includes |
