Posted Mon, 27 Jul 2015 12:03:42 GMT by Praveenkhanna Udayakumar
Hej Hej!!

I am Praveenkhanna from India. I am working on MIKE 21 Spectral Wave Modelling. I need some suggestions to reduce Simulation time.

Presently I have simulated the model setup for a domain of 1.8 deg x 1.5 deg with triangular mesh having 10710 no. of nodes and 20586 no. of elements and it takes 125 hours to complete one year simulation. I have no idea that how much time it will take for a similar setup to complete one year simulation.

Can anyone suggest me a better way to modify my mesh or spectral wave setup!!??

My system configuration is ,
2.4 GHz 8 core processor, 16 GB RAM, 64 bit operating system and Windows 8.1 Os.

Thanks and Regards
Praveenkhanna U
Posted Mon, 27 Jul 2015 18:55:59 GMT by Ricardo Machado Student
Hello Praveenkhanna,

There are several ways to decrease computation time:
1) Increase the simulation time step (but not so much that it causes instability in the model)
2) Decrease the number of mesh elements
3) Choose less computationally expensive formulations such as "directionally decoupled parametric formulation" and "quasi stationary formilation"
4) Use lower order, fast algorithm for the geographical space discretization
5) Increase the values of other time steps
6) Do not include effects which you believe will be of little relevance for your modelling purposes or for the particular case at hand
7)  etc etc.

There's a ton of parameters which can affect your simulation time and where to make changes really depends on your model.
I would suggest you try a very simple model at first and build its complexity as you get a feel for its behavior.

Best regards,
Ricardo Machado
Posted Tue, 28 Jul 2015 12:17:55 GMT by Praveenkhanna Udayakumar
Thanks for your reply Ricardo Machado.

Any more suggestions from other members???!!!!
Posted Wed, 05 Aug 2015 10:50:59 GMT by Murali Surisetty
[list]Hi Praveen,

Further to Ricardo's points, you need to consider spending some extra time generating the mesh bathymetry for the model. For your above mentioned domain nos of nodes/elements seem high.
[list type=decimal]
[li]Create you mesh in the multiples of 3 e.g. from boundary to area of interest 3000m^2, 1000m^2, 300m^2, 100m^2,30m^2 resolution mesh. [/li]
[li]Check you land boundary file to eliminate closely grouped nodes/vertices. Avoid clusters of  vertices especially in areas which are of least importance for your study.
[/li][li]Load and Analyse the mesh after exporting the interpolated mesh based on "Smallest time step based on the water depth" and edit the troubling nodes.(this alone would save up-to min 15% of simulation time)
[/li][/list]

Hope these tips will be helpful to you.

Best regards,
Murali
[/list]
Posted Wed, 05 Aug 2015 12:19:44 GMT by Praveenkhanna Udayakumar
Thanks for your Reply Mr.Murali.
Actually I tried every method possible.
My query is mainly to know how much time takes to complete one year simulation (domain of 200km x 200km, nearshore mesh resolution is of ~3000m^2 m offshore is of about ~50000m^2). My Model setup is taking about 5 days to complete. Is it normal or slow?

Regards
Praveenkhanna
Posted Tue, 11 Aug 2015 12:46:43 GMT by Ricardo Machado Student
It's hard to say whether it is normal or slow. It highly depends on the computer you are using. But if you can't afford to wait that long you will just have to simplify your model.

Best regards,
Ricardo Machado
Posted Mon, 17 Aug 2015 04:20:46 GMT by Praveenkhanna Udayakumar
Thanks for your reply Mr. Ricardo Machado...

-Praveenkhanna

You must be signed in to post in this forum.