Advanced Tips for Using Flux and Slurm in High-Performance Computing
Explore helpful tips and best practices for maximizing resource management efficiency with Flux and Slurm in high-performance computing environments. Learn about exclusive resource allocation, mpibind optimization, job scheduling strategies, and more.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
Flux (and a bit of Slurm) LC User Meeting March 2025 Ryan Day, LC Resource Management Prepared by LLNL under Contract DE-AC52-07NA27344. 1 LLNL-PRES-2003413
Documentation https://hpc.llnl.gov/banks-allocations-and-jobs https://hpc-tutorials.llnl.gov/flux/ https://flux-framework.readthedocs.io/en/latest/ https://github.com/LLNL/mpibind/blob/master/tutorials/flux/mod ule1.md https://hpc.llnl.gov/documentation/tutorials/using-lc-s-sierra- systems `man flux-COMMAND`, `flux -h`, `flux COMMAND -h` 2 LLNL-PRES-2003413
--exclusive and mpibind `--exclusive` is weird in Slurm. `salloc | sbatch | srun (outside of an allocation)`: all of the resources. `srun (inside of an allocation)`: minimal resource set. `srun -N1 -n4 --exclusive -p pdebug my_program`: probably won t do what you want. `salloc -N1 --exclusive -p pdebug srun -n4 my_program`: for 4 tasks that see the whole node. Flux defaults to giving a minimal resource set. `--exclusive`: always gives you all of the resources on the node(s). Mpibindworks best with all of the resources. `flux run -N1 -n4 my_program`: will only see one GPU on elcap. `flux run -N1 -n4 --exclusive my_program`: will see all 4 GPUs. 3 LLNL-PRES-2003413
Other Flux notes for ElCap and Tuolumne `-b standby`: jobs automatically killed if doing so would allow highest priority job to run. `flux update JOBID duration= `: users can shorten their jobs. LC Staff can extend them. `flux urgency JOBID `: users can hold/release their jobs. LC Staff can expedite them. LC Staff can create limited access queues for DATs. SPINDLE is on by default. `-o spindle.level=off` or `-o fastload` to turn it off. Mpibindwill try to avoid noise reduction / system cores unless you `unset MPIBIND_RESTRICT`. If using `--setattr=gpu_mode= `, also set `--conf=resource.rediscover=true`. 4 LLNL-PRES-2003413
Documentation https://hpc.llnl.gov/banks-allocations-and-jobs https://hpc-tutorials.llnl.gov/flux/ https://flux-framework.readthedocs.io/en/latest/ https://github.com/LLNL/mpibind/blob/master/tutorials/flux/mod ule1.md https://hpc.llnl.gov/documentation/tutorials/using-lc-s-sierra- systems `man flux-COMMAND`, `flux -h`, `flux COMMAND -h` 5 LLNL-PRES-2003413
6 LLNL-PRES-2003413
