Job Submission changes¶
Correctness fixes¶
- The OpenMP example no longer overflows
intduring multiplication or signed 64-bit storage for the final sum. - The OpenMP request starts at eight CPUs instead of assuming all 128 hardware threads are useful or configured as Slurm CPUs.
- MPI uses a coherent wrapper/module stack and a small two-node request;
srunis presented with a site-integration caveat. - Job IDs are opaque rather than assumed to have four or five digits.
- Early completion and wall-time termination are explained correctly.
- Pending behavior is tied to Slurm reason codes rather than described as universally indefinite.
--exclusiveis limited to genuine whole-node cases.- GPU toolkit and driver capability are distinguished.
Usability additions¶
Added a resource-model table, safe shell baseline, job arrays, accounting-driven tuning and a support checklist. Module and partition placeholders avoid freezing a historical software version into a supposedly current template.
Full diff¶
Loading diff…