Using the commands shown in the previous tutorial, list your jobs from the past month and print out enough details so that you can remember the purpose of the jobs (perhaps job ID, the name, partition, start time, allocated cores, used and requested memory). The job ID could be connected to Slurm output files, if you still have those available.
Some tips:
sacct
command is a rather heavy query to the Slurm accounting database.
Don’t make large queries for testing as we don’t want to slow down jobs getting
scheduled to run! Start with the past day or so until you get the syntax right.
This applies to learning all new commands and applications.man sacct
you can see which keywords to use for printing out different
fields stored in the database.sacct -S 2022-11-01 > sacct_output.txt
more
or less
) to look
at the contents (and grep
, awk
, python
, etc. to extract/analyze data).Some things to look for:
sacct
output parameters
you should be able to find representative jobs for each type that you’ve been
running recently) and use seff
to look for the CPU Efficiency.seff
)Use this information to set the resource requests for similar new jobs. You can also look in My CSC for previous (monthly) usage per project.