[Wrf-users] test cases and run parameters for I/O study

[Wei-keng Liao]

Hi, Saji

From the WRF test cases I ran, I am seeing one read operation at the beginning
and several writes after, depending on the run time and history_interval.
In your case, it sounds like read operations can also be multiple. I am very
interested in running a realistic case and very appreciated you are willing to
provide me an example data/setting. I will try build WPS and let you know.

By the way, I have read a few papers that mentioned I/O in WRF can be a bottleneck
even if PnetCDF or I/O quilting is used. As a PnetCDF developer, I would like
to check if it is something in PnetCDF or in WRF that causes the problem.
Would you mind letting me know whether I/O performance is ever a problem to you?

best

Wei-keng

On Jun 19, 2014, at 3:11 AM, Saji Hameed wrote:

> Hi Wei-keng,
> 
> I think you are doing it the right way. However the I/O part is mostly the Output in this idealized case.
> 
> If you need a more realistic example, complete with realistic input, I could provide you one from
> our operational forecasts or simulations. We try to limit the output size, but you can change
> the configuration to create huge output files.  You may also need to know how to run the
> WRF pre-processing software. 
> 
> Let me know.
> 
> cheers,
> saji
> 
> 
> Saji N Hameed,
> ARC-ENV, Center for Advanced Information Science and Technology,
> University of Aizu, Tsuruga, Ikki-machi,
> Aizuwakamatsu-shi, Fukushima 965-8580,
> Japan
>         
> Tel: +81242 37-2736
> Fax:+81242 37-2760
>  email: saji at u-aizu.ac.jp
>  url: http://enformtk.u-aizu.ac.jp
>  bib: http://www.researcherid.com/rid/B-9188-2009
> code: https://github.com/sajinh
> 
> 
> On Wed, Jun 18, 2014 at 7:29 AM, Wei-keng Liao <wkliao at eecs.northwestern.edu> wrote:
> Hi,
> 
> I am studying WRF's I/O pattern and wondering if I can get some
> recommendations on choosing a test case and its run-time parameters.
> The ideal parameter setting is that can produce a large domain size
> (I/O amount) and a short run time (with 3-5 output time snapshots).
> 
> I tried the test case "test/em_tropical_cyclone" and changed the
> following parameters in file namelist.input.
>  run_days                            = 0,
>  run_hours                           = 0,
>  run_minutes                         = 10,
>  history_interval                    = 5,
>  s_we                                = 1,
>  e_we                                = 401,
>  s_sn                                = 1,
>  e_sn                                = 401,
>  s_vert                              = 1,
>  e_vert                              = 121,
> 
> The above settings took about 10 minutes and produced an output file
> of size 3.3GB (with 3 time snapshots). I wonder if I am on the right
> track and will very appreciated if the experts in the WRF community
> can kindly provide me additional suggestions.
> 
> Wei-keng Liao
> 
> _______________________________________________
> Wrf-users mailing list
> Wrf-users at ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/wrf-users
>