Memory Usage

user_8673 · August 13, 2022, 11:04pm

Hi,

Attempting to couple TOMAWAC and TELEMAC-2D over a relatively large domain (600,000 elements) and noticed that [name removed] seems extremely memory hungry compared to T2D. Where this size of grid would not normally be an issue with T2D, TOMAWAC is using my entire 64 GB of RAM running standalone. Moreover, it is running extremely slow.

Currently trying to debug the simulation to make sure its not related to my setup but is this type of memory usage normal for TOMAWAC?

[name removed]

user_8473 · August 23, 2022, 8:49am

Hello,

You have to keep in mind one big different between telemac2d and tomawac.

For each point of your mesh,

telemac2d mainly stores 3 variables. h,u,v
tomawac stores the nf x ndire f unknown where nf is the number of frequencies and ndire the number of directions. so if for example you have 36 directions and 30 frequencies, tomawac stores around 300 hundred times plus than telemac2D.

So if possible you can try to reduce one of the two variables nf or ndire.

Or you can try to use the coupling that uses different meshes for telemac2d and tomawac. Tomawac does not need meshes as fines as telemac2d. In order to do so use tel2tom coupling with tomawac2 as keyword COUPLING WITH.

Examples can be seen in examples/sisyphe/littoral and examples/gaia/littoral-t2d-tom/

kind regards

user_8673 · August 29, 2022, 2:33pm

Hi,

That makes sense - thanks for the suggestions.

Are you aware if the TEL2TOM functionality has been extended to T3D as of yet? I actually need to run the the hydrodynamics with T3d eventually.

Regards,

user_8473 · August 29, 2022, 3:49pm

Hello

Yes [name removed] extended tel2tom to T3D, it is pushed in the main and should be in next version.

[name removed]

user_8673 · August 30, 2022, 7:08pm

Ok that’s great news.

Unfortunately [name removed] is far too slow for my particular application at the moment. I need to model a large basin with locally refined elements around coral reefs so am limited to a low timestep to resolve breaking. The memory usage is crazy even with reduced elements and spectral directions and frequencies.

I understand that models like SWAN are based on different numerical schemes and may be subject to increased error compared to TOMAWAC due to their implicit nature, but is there any discussion around trying to offer similar numerics or improvements to the TOMWAC? Being a long-term user of TELEMAC I’m always running into a wall when it comes to needing to use TOMWAC.

user_14856 · August 31, 2024, 9:54am

Hi, I’d like to have a clearer idea for Memory Usage.

I have a global mesh of the oceans of 6km resolution with

1.538.748 nodes and 3.006.764 elements.

I have a RAM of 64GB

When I launch TOMAWAC, I get 12.5% usage i.e. 8GB

It roughly corresponds to:

nodes x freq (24) x directions(31) x 8 bites = 9.1GB

It seems to make sense.

When I launch TELEMAC2D, I get 5% RAM usage i.e. 3.2GB

If I calculate as [name removed] has specified:

nodes x 8 bites x nparameters, it means I have roughly 260 parameters?

When I launch TELEMAC3D, I get 50% RAM usage i.e. 32GB

Here I really don’t know how to translate this.

For TELEMAC, I’d expect the RAM usage actually depend on the Matrix/Numerical Scheme you use, correct?

Could someone provide estimate for TELEMAC2D/3D? Or point me towards the documentation, if there is a section on this aspect?

BTW, I’d had expected TOMAWAC to be more needy than TELEMAC3D, so this is not that bad after all IMO [name removed]