WRF Optimization Help Needed for High-Resolution MGW Simulation on Derecho

[ananyas]

Hello,

I am currently working on a case study focusing on a MGW event spanning March 6th to 7th, 2008, and I am running into a few challenges with the simulation slowness that I am hoping to get some advice on. For this study, I have configured my simulation with 3 nested domains, using a 1:4:4 parent-child ratio, where the third domain is set up at a high resolution of approximately 1.75 km. The simulation is set up to run on 32 CPUs on derecho.

Initially, the simulation is running for 1st and 2nd domains until the 18th hour but there is a significant slowdown starting from the 18th hour of March 6th (at the start of the third domain's processing). This slowdown progresses to the point where the simulation times out at the 22nd-hour mark, unable to complete within the allocated 6-hour wall-clock time limit (I cannot request a more than 6-hour time limit because of the constraint).

I was wondering if anybody might have any ideas on how to optimize the simulation - around domain configuration changes, namelist.input changes, try to request a higher wall clock time limit, etc.

Thanks in advance.

 

[kwerner]

Hi,
The namelist you attached is only set up for a single domain. If you are running on derecho, and if you don't mind, you can point us to the directory where you're running so we can access the files and take a look at them. If you'd rather just attach the files, please attach the correct namelist with all 3 domains, as well as your rsl* files (you can package those together in a single *.tar file and attach that). Thanks.