I have started to run a two way nesting simulation with a grid size of 1079*755 755*663, for running 2 minutes the time taken is 3 minutes approximately(slower as compared to standalone model), I am attaching my .in file for your reference.
Kindly suggest a solution for this one. I have seen previous threads on ths ame topic, however it dates back to previous version.
Slow running of nesting simulation
-
- Posts: 30
- Joined: Fri Dec 15, 2017 6:07 pm
Slow running of nesting simulation
- Attachments
-
ocean_india.in
- (118.94 KiB) Downloaded 230 times
Re: Slow running of nesting simulation
Yes, the online nesting is extremely slow. I have not seen any ROMS development in the myroms code to lead me to expect that this has changed since those old posts. I am instead doing offline nesting because it gives more flexibility in the child grids.
-
- Posts: 30
- Joined: Fri Dec 15, 2017 6:07 pm
Re: Slow running of nesting simulation
Thank you for the reply Kate. I tried with multiple combinations of Ntilei and NtileJ keeping their ratio almost equal to Lm/Mm, as I reduced the total number of processors from 40x30 to 10x7 (40/30=1.33~Lm/Mm)(10/7=1.423~Lm/Mm) I was able to decrease the time taken for running the model considerably. Now almost 4Hrs 20Min are completed in 32 minutes.
-
- Posts: 30
- Joined: Fri Dec 15, 2017 6:07 pm
Re: Slow running of nesting simulation
Hi kate,
I need your help to resolve one of my doubt regarding Nesting, I have a coarser grid (1/16) and a refined grid(1/48) the refined grid is created from the coarse using coarse2refine.m matlab script. Rather than having a higher resolution bathymetry for the refined grid, the same bathymetry is observed for the refined grid as compared to the coarser one. How could that happen. I suppose that you have also used the same method, if there is any difference in the approach please highlight the difference
I need your help to resolve one of my doubt regarding Nesting, I have a coarser grid (1/16) and a refined grid(1/48) the refined grid is created from the coarse using coarse2refine.m matlab script. Rather than having a higher resolution bathymetry for the refined grid, the same bathymetry is observed for the refined grid as compared to the coarser one. How could that happen. I suppose that you have also used the same method, if there is any difference in the approach please highlight the difference
Re: Slow running of nesting simulation
Hernan once said that he thought the way to do it is fine-to-coarse, with a fine grid the size of your larger domain. However, as you find, one must use coarse-to-fine because that's the matlab script you have. I made all the grids and did the bathymetry the way I wanted to.
What I can help you with is the roms.in file. If you have a working roms.in for the coarse grid and same for the fine grid, I have a python script to merge the two (or three, or five). See the pyroms/examples/nesting directory.
What I can help you with is the roms.in file. If you have a working roms.in for the coarse grid and same for the fine grid, I have a python script to merge the two (or three, or five). See the pyroms/examples/nesting directory.
-
- Posts: 30
- Joined: Fri Dec 15, 2017 6:07 pm
Re: Slow running of nesting simulation
My standalone ROMS set up for coarse(1/12) and fine(1/36) resolution is running well individually but when I use two way nesting the Cw value becomes greater than one for finer resolution grid. Is that problem due to the higher time step? I have used 300,100 for coarse and finer resolution respectively and have used the same dt for running independently too.