Research Support Handbook

FAQ

Why is my job running much longer than expected?

• Always check if your script is really benefitting from parallelism. Are you really using all the threads/cores you got allocated via slurm? (log in the compute node via ssh and check the processes running htop -u *vunetid*).
• Note that the cpu in high performance servers does not necessarily clocks faster than your laptop cpu, the real performance gain comes from parallelism.
• Is your dataset very big? Always use the local scratch space.
• Use profilers to assess your code for bottlenecks. (Python: cProfile, R: profvis, C++: perf)

Why does the software I installed via conda return unexpected errors?

Conda injects settings in your .bashrc file once initialized and prepends directories to your $PATH variables. This can result in dependency errors for software that is actually relying on native system dependencies.