Tips and tricks!?

[Saint]

Hello again,
I hope you don't mind me posting a new topic.

In the past, when I was using my old software on win7 and I had different tricks to boost the simulation or reduce the unnecessary task of the windows.
Now, I wonder if you have the same process?
If you are using windows, (preferably win 10), What do you do with heavy simulations?

Do you kill any service or process ? (in win 7, I would kill almost every extra service, even the main explore.exe, and I would set the priority of my simulation/solver/software to high). How about win 10? (Im new with it)

how about the internal setting of Elmer? is there a way to boost the speed, or even opposite, to reduce the pressure/speed, to prevent it from crashing?

My laptop has multiple cores, and its RAM and graphic is not bad, but a friend of mine which is using the same laptop, but lower model (cpu,ram...) has no problem with his simulations with ansys, I think he is effectively using more cores in his simulations.
I saw a parallel simulation option; but to be frank, I just hoped to work with it without reading about it first and I failed

Also, about the solver detail, not only the type, but even the subcategory of the solver: CG. SCG, BIGCGstabl ...
I don't dare to ask for guidance about them, since I know they are a complex world by themselves. But generally speaking, how effective they are in convergence and simulation speed; is there a prefered stup, for slowing down or speeding up the simulation ?

Wish you sunshiny days,

[Saint]

honestly, I was expecting a heated conversation for this topic. I believe it is safe to assume most people are using linux, therefore no one has any points to add.

Maybe, administrators might want to delete this topic; since there is no useful post in it.

[raback]

Hi

I think one issue here is that most of us use Elmer almost exclusively on Linux. This is also the natural platform for supercomputing so if you want to got to larger problems things will be almost the same.

The parallel computing of Elmer has been touched in webinar series, the slides are here
https://www.nic.funet.fi/pub/sci/physic ... puting.pdf

Currently we have a benchmarking initiative where the aim is to make it easier to find optimal solver settings for different problems, see:
https://github.com/ElmerCSC/elmer-linsys

The devil always lies in the detail. So different problems have different optimal solutions. For example, if you are solving magnetostatics you cannot use ILUn as preconditioner because the vector potential is not uniquely defined (only its curl is) whereas for Poisson equation this is usually very good strategy. For vector valued problems like the Navier or Stokes equations the block preconditioning is often hard to beat. So the tips very much depend on what kind of problem you're about to solve.

-Peter

[Saint]

Hi

I think one issue here is that most of us use Elmer almost exclusively on Linux. This is also the natural platform for supercomputing so if you want to got to larger problems things will be almost the same.

The parallel computing of Elmer has been touched in webinar series, the slides are here
https://www.nic.funet.fi/pub/sci/physic ... puting.pdf

Currently we have a benchmarking initiative where the aim is to make it easier to find optimal solver settings for different problems, see:
https://github.com/ElmerCSC/elmer-linsys

The devil always lies in the detail. So different problems have different optimal solutions. For example, if you are solving magnetostatics you cannot use ILUn as preconditioner because the vector potential is not uniquely defined (only its curl is) whereas for Poisson equation this is usually very good strategy. For vector valued problems like the Navier or Stokes equations the block preconditioning is often hard to beat. So the tips very much depend on what kind of problem you're about to solve.

-Peter

Thank you so much for the PDF, interesting ! I am trying to smooth out my graph, and I thought, modifying the solver detail, will be right way, but I had no idea about the effect and functionality of them.... unfortunately, I don't have enough time and resource to learn Elmer or even CFD in a proper way.
Lack, of job and visa dealine, forces me to be results oriented, instead of learning it fundamentaly.

about the benchmarking, i have no idea how to use it, github confuses me, can you explain it how I can read or learn fro it ?