| | 1 | = Job-level scheduler customization = |
| | 2 | |
| | 3 | sched/sched_customize.cpp contains the following function: |
| | 4 | {{{ |
| | 5 | bool wu_is_infeasible_custom(WORKUNIT& wu, APP& app, BEST_APP_VERSION& bav); |
| | 6 | }}} |
| | 7 | |
| | 8 | This is called for every job that the scheduler |
| | 9 | considers dispatching. |
| | 10 | If it return true, the scheduler won't send the job to this host. |
| | 11 | This allows you to make project-specific scheduling decisions on a per-job basis. |
| | 12 | |
| | 13 | The default implementation simply returns false. |
| | 14 | However, you can customize it. |
| | 15 | The source code contains (commented out) the following possible policies: |
| | 16 | |
| | 17 | * Don't use GPU versions for particular jobs. |
| | 18 | * Require a certain number of GPU processors, as specified by WU.batch. |
| | 19 | * Reduce the FLOPS estimate for particular jobs. |
