open TELEMAC-MASCARET The mathematically superior suite of solvers

Hi

I'm working with binary atmospheric files in Telemac2D and 3D in v8p1r2. The models work well, files are read and the model runs as expected.

However, I was wondering if anyone has experienced what I'm running into. At the start of the simulation, the model runs blistering fast, but after almost two weeks of simulation time, the model starts to grind to a slow crawl. The model is still running and the number of iterations remain small implying its still stable, just the time per iteration has grown considerably. Checking the outputs afterwards everything is stable.

For example, I currently have a simulation at approximately 15 days and one that just started. In the time the first one has progressed two hours, the second simulation has progressed almost six days.

If anyone has any insight or if its just a wait it out moment. But I am experiencing this with multiple different model domains and the last two versions of Telemac. Don't know if its a memory allocation or something?

Many thanks

David

I should also add, that when I use a spatially constant but time varying atmospheric file (ascii file), the model runs at the same pace for the entire run and completes with no issue. This implies its something specific to the binary atmospheric file.

David

Hi David

I already seen this problem and yes, you're right, it's only for binary atmospheric file.

My understanding of the problem is you're always accessing to the binary file to read the values at before and after the time of the computation to perform the interpolation in time.
When values are available each 1 or 3 hours and the computation is done every 5s this is a loss of time.
I suspect also a kind of longer time to access to later time in the end of atmospheric file but not fully sure...
To sort out this problem I remember I make a small change to just recompute the interpolation at the right time without looking for the surrounding time in the file. This means to store the values before and after (with time of course) and change this value only if the time request this change

Hope this will help as I'm not sure I could easily find the fortran of my personal solution...

Many thanks Christophe, I will investigate your suggestion. Glad it wasn't just me going mad!

David

Hi guys,

Also running into this problem - very noticeable with both atmospheric files (winds and pressure) and wave radiation forces being interpolated on a very high resolution mesh.

Attempting to modify the source at the moment as Christophe suggested but my FORTRAN sills are limited to say the least and having difficulty figuring out a way to make the code universal for any variable of interest.

By any chance has anyone made the required subroutine changes and would be willing to share? If this could be implemented into the trunk it'd be a great improvement to the computation speed.

Regards,
Toby

Hi Toby

Unfortunately I was unable to complete any successful modifications.

Indeed if anyone has successfully implemented a solution I would be very interested in it becoming incorporated into the trunk.

Kind regards

David

Hello David and Toby,

Can you send your steering files (and possible Fortran files) so that we can see in which subroutines you pass please?
And also the release you use.

Chi-Tuan

Hi Chi-Tuan,

Haven't had a chance to look into this again until now. I am experiencing the issue on a model I am currently working on though. Simulation is rapid for the first few days and then becomes very slow.

I can confirm this is due to the input files as the simulations are batched by month using a restart file for the last timestep of the previous month. As the next month commences, the simulation returns to being relatively fast.

This is confidential project so I am unable to share the files publicly but have sent you in a private message if you'd like to investigate.

Regards,
Toby

Seems I'm unable to send a private message for whatever reason.

Please feel free to reach out to me privately on the forums (I'm unable to figure it out) and I would be more than happy to send the files through.

Regards,
Toby

Dear Toby, David and Christophe,

Please find enclosed 2 modifications for METEO subroutine (in sources/telemac2d) and METEO_TELEMAC module (in sources/utils/bief).

I have tested them for a configuration sent by Toby (many thanks) and succeeded in getting exactly same results as using METEO_FROM_BINARY_FILE and faster (up to twice faster depending on the number of cores for this case).
Yet, there is a need to fit expected binary format for METEO_TELEMAC with the name of variables (as described in the WAQTEL user manual, the name of variables for wind and atmospheric pressure have to be WINDX, WINDY and PATM (but can also be WINDS, WINDD for wind velocity magnitude and direction + PATM).

You can change the name of the variables in your BINARY ATMOSPHERIC DATA FILE by running run_telfile.py alter --rename 'old name var=new name var' one by one, e.g.

Please give me your feedback if possible before September 10th so that it can be merged to branch v8p5.

Hope this helps,

Chi-Tuan