WRF-Fire merge testing

From openwfm
Revision as of 08:46, 10 February 2019 by Afarguell (talk | contribs) (→‎Intel)
Jump to navigation Jump to search


Please see Talk:WRF-Fire_merge_testing for permanent notes.

The basics

The version to test is branch fuel-moisture-model currently at commit c1cedfc7eba28a2ed

The baseline is WRF4 branch develop currently at commit 0f296f9c0f674f9

When either branch advances we have to retest - after any fix in fuel-moisture-model, and, after we are done, we will have to forward branch develop to current WRF4 and merge it into fuel-moisture-model. For this reason (and to reduce confusion), I am trying to organize the tests in a way that they are really easy to rerun without modifying any files - separate clones for different compilation options and separate copies of test/em_fire directory for different test cases.

The namelist.input and namelist.fire with variables that turn the fuel moisture model and fire on are in test/em_fire_rain/rain. The description of the namelist variables is at Fuel_moisture_model#Configuration. Other features from WRF-SFIRE are not carried over - only the fuel moisture model.

Who is doing what

  • Angel - WTF, parallel testing of both branches for identical results, all test cases in test/em_fire
  • Adam - a real problem
  • Jan - fixing, selected testing as needed


How to test

  • Record also unsuccessful tests, when fixed, replace by the final result and include link to the commit used
  • All tests should be reproducible by checking out the commit noted, which should include all data for ideal runs
  • For real runs, please link to the datasets to be reproducible

What needs to be tested

  • Fuel moisture model off vs. WRF4 baseline, with and without fire, did we change anything?
  • reading fmc_g, is it doing what it should?
  • Fuel moisture model on, fire on, is it doing what it should?
  • Serial and parallel execution with different numbers of CPUs, are the numbers same?
  • WTF as specified in WRF testing requirements
  • All ideal cases in test/em_fire and some real

Jan's test results

All kingspeak runs are in /uufs/chpc.utah.edu/common/home/u6015690/merge and run on node kingspeak11 (nodes can be different).

Baseline

branch develop 0f296f9c0f674f9

  • ifort serial (13), nesting 0, configure -d, WRF-Fire-merge-develop/test/em_fire/

Using fmc_g

Fuel moisture model

branch fuel-moisture-model, copies of test/em_fire as em_fire_xxxx

Angel's test results

All kingspeak runs are in /uufs/chpc.utah.edu/common/home/kochanski-group3/farguella/merge.

Branch develop vs. added-fmc_g

The branch develop and branch added-fmc_g are in kingspeak in folders WRF-Fire-merge-develop and WRF-Fire-merge-added-fmc_g respectively. They are tested in order to see the first new functionalities of adding variable FMC_G in WRF4 baseline. Tests are from list in https://www.openwfm.org/index.php?title=Porting_WRF-SFIRE_fuel_moisture_model_to_WRF4#Testing_to_be_done.

All the runs are on node kingspeak11 and kinspeak12 and all the compilations are done using ifort serial (13), nesting 0.

fire_fmc_read not present in namelist.input

One can observe that both simple hill simulations are exactly the same running 

diff WRF-Fire-merge-develop/test/em_fire/wrf.log WRF-Fire-merge-added-fmc_g/test/em_fire/hill_tests/original/wrf.log -I Timing

which returns 

10c10
<    alloc_space_field: domain            1 ,              344571608  bytes allocated
---
>    alloc_space_field: domain            1 ,              345388824  bytes allocated

Therefore, the only difference is in the allocation size as expected.

However, the new branch added-fmc_g is creating new variable FMC_G taken constant value of fuelmc_g from namelist.fire all over the domain. Changing value of fuelmc_g in namelist.fire changes the value of the new variable FMC_G.

fire_fmc_read present in namelist.input

Setting 

fire_fmc_read=0

and poblating FMC_G variable in wrfinput_d01 using ncreplace in Matlab. The branch added-fmc_g uses the new FMC_G in the simulation from new wrfinputs. Tested using:

  • Constant FMC_G field of 0.07: WRF-Fire-merge-added-fmc_g/test/em_fire/hill_tests/fmc_07.
  • Constant FMC_G field of 0.08: WRF-Fire-merge-added-fmc_g/test/em_fire/hill_tests/fmc_08. Exactly the same results that branch develop which was taking the information from namelist.fire file.
  • Random FMC_G field: WRF-Fire-merge-added-fmc_g/test/em_fire/hill_tests/fmc_rand. Gives the progression with random bumps as expected.
  • Slope FMC_G field: RF-Fire-merge-added-fmc_g/test/em_fire/hill_tests/fmc_slope.

Branch develop vs. fuel-moisture-model without fuel moisture

The experiments tested are hill_simple and two_fires ideal cases. In the branch fuel-moisture-model, the fuel moisture model is turned off and we want to obtain exactly the same results than using branch develop.

The branch develop and branch fuel-moisture-model are compiled and runned in Cheyenne in folders /gpfs/fs1/p/univ/ucud0004/angelfc/merge/develop* and /gpfs/fs1/p/univ/ucud0004/angelfc/merge/fuel-moiture-model* respectively.

All the simulations are done in folders test/em_fire/hill_simple and test/em_fire/two_fires.

Intel

  • Serial executions: develop-intel-serial and fuel-moisture-model-intel-serial. They are compiled using configure option: ifort compiler with icc serial (13), nesting 0. Both hill simple and two fires cases give exactly the same results.
  • OpenMP executions: develop-intel-openmp and fuel-moisture-model-intel-openmp. They are compiled using configure option: ifort compiler with icc dmpar (14), nesting 0.
  • MPI executions: develop-intel-mpi and fuel-moisture-model-intel-mpi. They are compiled using configure option: ifort compiler with icc dmpar (15), nesting 1. In both ideal cases hill simple and two fires, same results in variables U, V, and TIGN_G from wrfouts. Using 1, 2, 4 and 8 MPI processes.
  • Hybrid executions: develop-intel-hybrid and fuel-moisture-model-intel-hybrid. They are compiled using configure option: ifort compiler with icc dmpar (16), nesting 1. All the executions crash because of memory corruption.

PGI

GNU

WRF in fuel-moisture-model branch

In order to test WRF in the new branch fuel-moisture-model, the WTF is used.

Running WTF in Cheyenne

The new WTF code is downloaded using 

wget http://www2.mmm.ucar.edu/wrf/src/wtf_v04.09_small.tar

After that, the tar file is uncompressed using 

tar -xvf wtf_v04.09_small.tar
cd WTF_v04.09

Once one try to use ./run_from_github.py python execution, an error appear which says that Data folder is old. So, the new Data folder is downloaded using 

cd ..
wget http://www2.mmm.ucar.edu/wrf/tmp/data_v04.08.tar

After that, the new data is uncompressed and replaced using 

tar -xvf data_v04.08.tar
rm -fr WTF_v04.09/Data
mv Data WTF_v04.09

Then, the new version of the data is updated in line 42 of run_from_github.py file as 

version="v04.08"

It is also needed to provide the account code to run the qsub in file qsub_this_to_run_all_compiler.csh. In our case, it should be 

#PBS -A UCUD0004

Finally, one can run the code doing 

./run_from_github.py

One needs to give also the github repository to test and the branch which are going to be: 

URL of the repository: https://github.com/openwfm/WRF-Fire-merge
branch: fuel-moisture-model

Results

Adam's test results

See also

Retrieved from "http://wiki.openwfm.org/mediawiki/index.php?title=WRF-Fire_merge_testing&oldid=3445"