Finland case (12 Aug 2017)
Report for phase 1
Reports and results are found here:
- Validation between baseline and prototype model runs, Nov 2023
- MetView instructions on how to plot
- Overview of runs 2022-2023 and verification
- Validation with radar
- Validation of winds
- Planned runs 2023, status
Oct 2023
The Finland_2017 case has been run with new prototype model (tagged version v0.1.0) and the outcome/validation is available in the link above "Validation between baseline and prototype model runs, Nov 2023".
It was seen that the decomposition did not work properly, using prototype model. And as a consequence using more nodes/processors did not speed up the run-time.
I have done a few tests setting NPRGPNS, NPRGPEW, NPRTRW and NPRTRV.
I hardcoded the numbers in namelist: deode/namelist_generation_input/CY46h1/master_namelists.yml into section "f4_common" and the "NAMPAR0" part.
But I assume one could use the following syntax to set the variables:
NPRGPNS = NPRTRW (=NPRTRNS) -> "NPROCY"
NPRGPEW = NPRTRV (=NPRTRN) -> "NPROCX"
| Case | NPROC | NTASKS | NODES | NPRGPNS | NPRGPEW | NPRTRW | NPRTRV | OMP_NUM_THREADS | STEP-TIME | RUN-TIME | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Ref | 2304 | 2304 | 18 | -- | -- | -- | -- | 1 | 1.3-1.6 | 4h | 15min |
| Decomp1 | 2304 | 2304 | 18 | 48 | 48 | 48 | 48 | 1 | 1.2-1.3 | 3h | 50min |
| Decomp2 | 2304 | 2304 | 18 | 48 | 48 | 48 | 48 | 8 | Crash... | ||
| Ref | 3840 | 3840 | 30 | -- | -- | -- | -- | 1 | 0.9-1.5 | 4h | |
| Decomp3 | 3840 | 3840 | 30 | 64 | 60 | 64 | 60 | 1 | 0.6-0.7 | 2h | 50min |
| Decomp4 | 3840 | 3840 | 30 | 64 | 60 | 64 | 60 | 8 | Crash... |
Decomp1 log-file: /home/fie/deode_ecflow/job/Decomp1/20200629/0000/Cycle/Forecasting/Forecast.1
Decomp3 log-file: /home/fie/deode_ecflow/job/Decomp3/20200629/0000/Cycle/Forecasting/Forecast.1
There is an increase in speed-up with these new settings.
NEW: cy46-deode branch for Finland case (12 Aug 2017): MARS data, with Satellite, 1 month spin-up (Erik Gregow)
- Status of progress 14 March 2023: DONE
- Better results, compared to not using satellite data!!
- archived datasets on (ECFS location: ec:/fie/harmonie/cy46h1_EDT_ref_fin/2017/08)
- 23 Feb 2023
- Code: https://github.com/uandrae/vhr_ws_scalability_exercise
- Testing the default setup of different parameter choices -> Works mainly fine! Some issues with the default options, gives crashes that I'm not sure why... Ask Ulf!
- Example of run-time cost for my 200m resolution domain, with single vs double precision:
-
Status of progress
- Status 15 Nov 2022: Run has started, running spin-up period
- Status 21 Nov 2022: Still running spin-up period (takes time with MARS... DTG: 29 Jul 2017)
- Status 28 Nov 2022: Spin-up done! Started run for case.
- Status 13 Dec 2022: Case-period soon done (running last day of fc)...
- Status 14 Dec 2022: Case-period finished
-
Using Cy43-deode branch, run by Erik Gregow
-
Start with one month spin-up
-
Data available under between 11 and 14 Aug 2017, hourly forecast with leadtime of 48h
-
Local Exp setup: /home/fie/hm_home/cy43_deode_ref_fin/
-
Source code: /perm/fie/source_code_exp/cy43_prod_deode/
-
archived datasets on (ECFS location: ec:/fie/harmonie/cy43_deode_ref_fin/2017/08)
-
Estimated storage at ECFS:
- Case period 10-13 Aug (+48h fc output): 2.5 TB
- Spin-up period, 1 month: 8.0 TB
---------------------------------------***-----------------------------------------
- Status of progress: DONE
- archived datasets on (ECFS location: ec:/fie/harmonie/cy46h1_deode_ref_fin/2017/08)
TEST CY43-deode with down-scaling: MetCoOp-ARCUS data, HERCULES 750m domain (S. Finland), test-period: 2022081600-2022081800
- Status 15 Nov 2022: Test-run started and completed, results not analyzed
- Continue the work in WP5.4. See information on this WP5.4 page-link
29 Nov 2022:
** SOLVED: Restart the failing exp.s with "Harmonie prod". **
Crashes in mainly all "LogProgress".
Example: /perm/fie/HARMONIE/cy46h1_deode_down750_fin/Postprocessing/Hour/LogProgress.1
Something changed on Atos system?
Lmod has detected the following error: Could not find installation path for
_module gcc/8.4.1 _
While processing the following module(s):
Module fullname Module Filename
--------------- ---------------
gcc/8.4.1 /usr/local/apps/modulefiles/lmod/compilers/gcc/8.4.1.lua
ping server(ecflow-gen-fie-001:3141) succeeded in 00:00:00.004534 ~4 milliseconds
Lmod has detected the following error: Could not find installation path for
_module gcc/8.4.1 _
While processing the following module(s):
Module fullname Module Filename
--------------- ---------------
gcc/8.4.1 /usr/local/apps/modulefiles/lmod/compilers/gcc/8.4.1.lua
prgenv/gnu /usr/local/apps/modulefiles/lmod/prgenvs/prgenv/gnu.lua