Scheduled Downtime
On Tuesday 24 October 2023 @ 5pm MT the forums will be in read only mode in preparation for the downtime. On Wednesday 25 October 2023 @ 5am MT, this website will be down for maintenance and expected to return online later in the morning.
Normal Operations
The forums are back online with normal operations. If you notice any issues or errors related to the forums, please reach out to help@ucar.edu

3 General questions about CESM

wvsi3w

wvsi3w
Member
Hello,
I searched on the forum and couldn't find the answer to my following questions maybe some of you dear scientists would know the answer:

1- When I do "more rpointer.atm" on any simulation (either finished or still ongoing simulation) I see this:
spinup1degAD_Narval.datm.r.0885-01-01-00000.nc

spinup1degAD_Narval.datm.rs1.0885-01-01-00000.bin
Which I am not sure why there is a bin at the end of the second line. I checked with other rest directories online (for example WACCM6-TSMLT-GEO) and the rpointer for atm doesn't have that bin.

2- When you submit a job and do sq to check there is one part in the header that says "MIN_MEM" and for one system (Beluga cluster of Canada) this MIN_MEM is 92G (which I think I assigned it somewhere, but in my other system (Narval cluster of Canada) it is 0. I was wondering if this is important and if I must modify it somehow. The reason I am asking is because my Beluga simulations are very slow and I was thinking maybe that's because I assigned that 92G somewhere in the configuration of the model, on the other hand, I guess MIN_MEM is only referring to bash memory and maybe it is not relevant to the speed of run on beluga. I don't know. If you think this memory thing is important and I have to increase it to speed up my runs and it can be modified, do you think I should put "mem=186G or mem=752G" which is stated on the description of the Canadian cluster? BTW, I have increased the number of cores (nodes) to its highest possible amount (512 on Narval and 320 on Beluga).

3- I have used Instantaneous 6-hourly data from the CCSM4 LME dataset and I was wondering if there is a rule to use average data in our simulations? I have read some threads that some people prefer average rather than instantaneous but I am not sure what the difference is here. I am asking this because I have prepared about 50TB of LME dataset for my project which all is instantaneous and I need to make sure by knowing your opinion on the difference of these types of data. I have explained this question in this thread (Guidance on Running with LME datasets with CESM2.1.3 (CLM5)) on Paleo forum which I am not getting any response yet, so please let me know your thoughts.

Thanks a lot.
 

slevis

Moderator
Staff member
I will attempt to answer #3:
- 6-hourly instantaneous output gives you the values of your history fields at the end of the last timestep of each 6-hourly period.
- 6-hourly average output represents the 6-hour mean of your history fields and time corresponds to the middle of each 6-hour period; the time variable does not explicitly state this except in some very new versions of CAM.
 
Vote Upvote 1 Downvote

fischer

CSEG and Liaisons
Staff member
I'll take on question #2:
It looks like to me that the MIN_MEM is used to request the amount of memory you want per node. I'm not sure what compset you're using, but 92G should be plenty. A WACCM compset might be happier with 186G. Looking at the "Node characteristics" for Beluga, I would go with 186G all of the time since it has the most nodes available. You shouldn't see any benefit going above 186G. Your 320 node run on Beluga probably ran into scaling issues causing it to run slower.

Chris
 
Vote Upvote 1 Downvote

wvsi3w

wvsi3w
Member
I will attempt to answer #3:
- 6-hourly instantaneous output gives you the values of your history fields at the end of the last timestep of each 6-hourly period.
- 6-hourly average output represents the 6-hour mean of your history fields and time corresponds to the middle of each 6-hour period; the time variable does not explicitly state this except in some very new versions of CAM.
Thank you Sam for your helpful answer. I was also able to ask one of my supervisors and I got the same response, which seems to be no issue with this type data.
Thanks again.


By any chance do you know what the difference is between the 007 and 003 LME datasets (I also mentioned it here)
 
Vote Upvote 0 Downvote

wvsi3w

wvsi3w
Member
I'll take on question #2:
It looks like to me that the MIN_MEM is used to request the amount of memory you want per node. I'm not sure what compset you're using, but 92G should be plenty. A WACCM compset might be happier with 186G. Looking at the "Node characteristics" for Beluga, I would go with 186G all of the time since it has the most nodes available. You shouldn't see any benefit going above 186G. Your 320 node run on Beluga probably ran into scaling issues causing it to run slower.

Chris
Thanks alot Chris.
I have tested both Beluga and Narval systems regarding the scalability issues and 8 nodes for both system was ideal (fastest and without much queue time).
So, if I want to change "<directive> --mem=92G </directive>" in my config_batch file do you think it would affect the output of simulations (currently running AD spin-up (for IHistClm50BgcCrop compset 1deg))? Or should I wait till this is finished and change the batch file to higher memory for ND spin-up (or transient runs maybe)???
 
Vote Upvote 0 Downvote
Top