Topic: OOFEM Performance

Hi,

and a happy, successful and healthy new year to everyone.

I've got a question regarding the OOFEM performance, see also the attached input and output files.
I am calculating a system using 8400 cctplate elements. Calculating time for this is almost 400seconds.

Since I often calculate systems with 10.000-15.000 elements (only 2D), so a better performance would be very interesting for me.

Is there any way to improve speed of the calculations exept using the parallel mode, e.g. using a different solver?

Thank u for your answers!

Dennis

Post's attachments

oofeminput.txt 771.77 kb, 4 downloads since 2018-01-10 

You don't have the permssions to download the attachments of this post.

Re: OOFEM Performance

And the output file...

I am using OOFEM 2.4, WIN 10 64bit, Intel i7-5600 @ 2.60GHz and 8,00GB RAM

Thank you!

Post's attachments

oofemoutput.zip 513.9 kb, 2 downloads since 2018-01-10 

You don't have the permssions to download the attachments of this post.

3

Re: OOFEM Performance

Hi Dennis,

the reason for poor performance is that you use default solver with no profile optimization.
If you turn on profile optimization (edit problem record - the third line in the input):

LinearStatic nsteps 1 profileopt 1

you can get significant speedup (on my system from 3 minutes to less than one second)
Additional gain can be obtained by using different solver or running the problem in parallel (refer to input manual), but for this case the profile optimization seems to be sufficient.

Re: OOFEM Performance

Perfect! Thank you for your answer!!

Re: OOFEM Performance

Hi, I think you can try it with DSS. I suggest you to compile it again with DSS. And there are some documents on web on how to use it.

Dennis wrote:

Hi,

and a happy, successful and healthy new year to everyone.

I've got a question regarding the OOFEM performance, see also the attached input and output files.
I am calculating a system using 8400 cctplate elements. Calculating time for this is almost 400seconds.

Since I often calculate systems with 10.000-15.000 elements (only 2D), so a better performance would be very interesting for me.

Is there any way to improve speed of the calculations exept using the parallel mode, e.g. using a different solver?

Thank u for your answers!

Dennis