-
Notifications
You must be signed in to change notification settings - Fork 7
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Feature/nl format #96
Conversation
change required cpu time and mem for global script
change required cpu time and mem
I changed the required cpu time and mem in the slurm script and readme based on my slurm settings on Hopper. Time might be a little off since I set ntask=12, but mem should be accurate (could be less depending on the case but it would be safer to put 60G). |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@angehung5 Please see my comments on your suggested memory updates here.
python/README.md
Outdated
#SBATCH --nodes=1 # Request N nodes | ||
#SBATCH --ntasks=1 # Request n tasks | ||
#SBATCH --mem-per-cpu=1000GB # Request nGB RAM per core | ||
#SBATCH --mem-per-cpu=60GB # Request nGB RAM per core |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@angehung5 A single node/task with your suggestion of 60GB memory may not work, please test this example.
I don't think we really need to increase ntasks because its not parallel code.
From my experience using a single node/task requires like about 256 GB memory.
run_canopy_slurm.sh
Outdated
#SBATCH --nodes=1 # Request N nodes | ||
#SBATCH --exclude=hop006,hop010,hop011 # Exclude some nodes (optional) | ||
#SBATCH --ntasks=1 # Request n tasks | ||
#SBATCH --mem-per-cpu=60GB # Request nGB RAM per core |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@angehung5 A single node/task with your suggestion of 60GB memory may not work, please test this example.
I don't think we really need to increase ntasks because its not parallel code.
From my experience using a single node/task requires like about 256 GB memory.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Tested on Hopper. For three time steps with node=1 and ntask=1:
Cores: 1
CPU Utilized: 01:12:51
CPU Efficiency: 95.37% of 01:16:23 core-walltime
Job Wall-clock time: 01:16:23
Memory Utilized: 9.57 GB
I put 12G just in case.
@angehung5 OK, thank you. Is this for running both the python global process script and canopy-app for those times? |
global process script only. |
@angehung5 Right, then the memory suggestion is not correct, and would need to be largely increased for canopy-app too. We need to test run both at same time/job. |
Closing this PR due to updated new PR from @angehung5 built on latest commit. |
@zmoon @angehung5 Since we have over 40 NL options and growing, need to organize better. I also added some python/README changes for running global data and an example slurm script to help. Please let me know ASAP what you think so we can update develop going forward.