Best Practices for HPC
1. Do NOT run any job which is longer than few minutes on the login nodes. Login node is for compilation of job. It is best to run the job on computes. (compute nodes)
2. It is recommended to refer the
quick start guide on the Paramshakti website. This should serve as a good starting point for the new users.
3. It is recommended to check the slurm sample scripts at /home/iitkgp/slurm-scripts for job submission.
4. Use the same compiler to compile different parts/modules/library-dependencies of an application. Using different compilers (e.g. pgcc + icc) to compile different parts of application may cause linking or execution issues.
5. Choosing appropriate compiler switches/flags/options (e.g. –O3) may increase the performance of application substantially (accuracy of output must be verified). Please refer to documentation of compilers (online / docs present inside compiler installation path / man pages etc.)
6. Modules/libraries used for execution should be the same as that used for compilations. This can be specified in the Job submission script.
7. Be aware of the amount of disk space utilized by your job(s). Do an estimate before submitting multiple jobs.
8. Please submit jobs only from $SCRATCH. You can back up your results/summaries in your $HOME
9. $SCRATCH is NOT backed up! Please download all your data to your Desktop/ Laptop.
10. Before installing any software in your home, ensure that it is from a reliable and safe source. Ransomware is on the rise!
11. Please do not use spaces while creating the directories and files.
12. Please inform PARAM Shakti support when you notice something strange - e.g. unexpected slowdowns, files missing/corrupted etc.
13. Please do not submit jobs to "gpu" or "gpu-low" partition unless your code strictly needs GPU based computing.
14. Please provide the wall-time parameter judiciously in job submission batch script which sgnificantly affects que waiting time.