Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.

Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

0% CPU but 100% RAM during "Assembling sparsity pattern"

Please login with a confirmed email address before reporting spam

After launching the study solving computation, COMSOL seems to behave weird.

I am stuck at 15% in the "Assembling sparsity pattern" process. 93% of my physical memory is taken by COMSOL (all 24GB of it!!!!) but 0% of my CPU is being used....which is rather weird to me. Is COMSOL doing anything at all or is "subtly" crashed?

If the CPU is not being used then surely the simulation is not running right? Or am I missing something?
Should I wait or should I stop it?

Thanks.

16 Replies Last Post 28 feb 2011, 12:12 GMT-5

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 25 feb 2011, 12:02 GMT-5
It depends. If the model is very large, your behavior make sense. What happens is that your solver is using massively the HD to store the data, and the writing-reading process is definitely slow, that's why you see your RAM completely full ("I wrote 24 GB and now I need something to write on") and your CPU at 0% (is relaxing while the HD is being written)...

It happened to me too...
It depends. If the model is very large, your behavior make sense. What happens is that your solver is using massively the HD to store the data, and the writing-reading process is definitely slow, that's why you see your RAM completely full ("I wrote 24 GB and now I need something to write on") and your CPU at 0% (is relaxing while the HD is being written)... It happened to me too...

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 25 feb 2011, 12:37 GMT-5
Ok,

Thanks for the input. I guess you are right since my 3D model has 4,000,000 elements.
(see Log in attachment)

I'll launch it and go for the week-end! :)

It would have been nice though to have some signals showing that the process is working and not just stuck (it was stuck at 15% for at least 1 hour).

Thanks again.
Ok, Thanks for the input. I guess you are right since my 3D model has 4,000,000 elements. (see Log in attachment) I'll launch it and go for the week-end! :) It would have been nice though to have some signals showing that the process is working and not just stuck (it was stuck at 15% for at least 1 hour). Thanks again.


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 25 feb 2011, 17:13 GMT-5
16 millions DoF? Is incredibly big, you should definitely try to reduce it. Seriously... It will take weeks to solve alredy in a "normal" way, with the swapping of data even more. I don't think you will ever get a solution with that model.
16 millions DoF? Is incredibly big, you should definitely try to reduce it. Seriously... It will take weeks to solve alredy in a "normal" way, with the swapping of data even more. I don't think you will ever get a solution with that model.

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 25 feb 2011, 17:20 GMT-5
That's why I first wanted to do that in 2D axisymmetric but COMSOL does not allow the use of cylindrical coordinate system in 2D axisymmetric (my curved layers have orthotropic properties following the curvature).

I already did the simulation in 2D but it's less representative of a real dome.

In 2D I used a cylindrical coord. system to map the material properties on the layers and in 3D I'm using a spherical coord. system.
That's why I first wanted to do that in 2D axisymmetric but COMSOL does not allow the use of cylindrical coordinate system in 2D axisymmetric (my curved layers have orthotropic properties following the curvature). I already did the simulation in 2D but it's less representative of a real dome. In 2D I used a cylindrical coord. system to map the material properties on the layers and in 3D I'm using a spherical coord. system.

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 25 feb 2011, 17:23 GMT-5
We should think at another solution, because with 16millions DoF you'll never get anything
We should think at another solution, because with 16millions DoF you'll never get anything

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 26 feb 2011, 06:34 GMT-5
Hi

But isnt 2D-axi cylindrical coordinates by default ? Or what else is it ?

Or was it spherical coordinates you were looking for ?

--
Good luck
Ivar
Hi But isnt 2D-axi cylindrical coordinates by default ? Or what else is it ? Or was it spherical coordinates you were looking for ? -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 26 feb 2011, 07:27 GMT-5
Hi Ivar,

I'm also working sometimes in 2D-axi and the coordinates are r and z, just like x and y. I don't honestly know if one can switch to cylindrical/polar/whatever...
Hi Ivar, I'm also working sometimes in 2D-axi and the coordinates are r and z, just like x and y. I don't honestly know if one can switch to cylindrical/polar/whatever...

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 26 feb 2011, 07:45 GMT-5
Hi

yes but "r and z" are cylindrical coordinates by design, no ?
with the axis at r=0 along Z

--
Good luck
Ivar
Hi yes but "r and z" are cylindrical coordinates by design, no ? with the axis at r=0 along Z -- Good luck Ivar

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 26 feb 2011, 07:51 GMT-5
Perhaps he needs the angle...
Perhaps he needs the angle...

Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 26 feb 2011, 10:04 GMT-5
Hi

which one ?

"phi" (0 <= phi =< 2*pi) the "loop angle" out of the paper is there (check the equations) but the physics is assumed to be independent from the phi angle.

I believe it's the spherical view he refers to, for which we need to add a spherical coordiante of the type:

rr = sqrt(r^2+z^2) rr>=0
psi = atan2(z,r) -pi/2 < psi < pi/2


--
Good luck
Ivar
Hi which one ? "phi" (0

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 28 feb 2011, 09:20 GMT-5
Hi,

Thanks for the discussion about the coord. ;)

2D-axi: Yes, the phi revolving around the z-axis is there but r is only a normal x-coord (that makes sense since for a cylinder).

No, what I'm looking for is creating a dome (think like a half empty sphere) made of layers of composites. My first attempt in 2D-axi was to create the arc of the dome, centered at (0,0) and revolving around z-axis to create the half empty sphere BUT there was no secondary coordinate system I could use to map my material properties along the arc of the sphere.

I tried to map a coordinate system by using:
r = (R^2+Z^2)^0.5
theta = arctan (Z/R)

The problem I get is that the theta equation is not accepted, saying that COMSOL is expecting something giving out a length [m] and not something giving out an angle.

Anyway, long story short, now I'm in 3D with my huge mesh...I reworked it and now I reduced from 4 millions elements to 1 million.

There might be some problems at one point due to low elements density between the composites plies but I can't do otherwise.
Hi, Thanks for the discussion about the coord. ;) 2D-axi: Yes, the phi revolving around the z-axis is there but r is only a normal x-coord (that makes sense since for a cylinder). No, what I'm looking for is creating a dome (think like a half empty sphere) made of layers of composites. My first attempt in 2D-axi was to create the arc of the dome, centered at (0,0) and revolving around z-axis to create the half empty sphere BUT there was no secondary coordinate system I could use to map my material properties along the arc of the sphere. I tried to map a coordinate system by using: r = (R^2+Z^2)^0.5 theta = arctan (Z/R) The problem I get is that the theta equation is not accepted, saying that COMSOL is expecting something giving out a length [m] and not something giving out an angle. Anyway, long story short, now I'm in 3D with my huge mesh...I reworked it and now I reduced from 4 millions elements to 1 million. There might be some problems at one point due to low elements density between the composites plies but I can't do otherwise.

Jeff Hiller COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 28 feb 2011, 09:46 GMT-5
Hello Dominique,
I am not sure how you mesh that hollow half sphere, but I can't see how you would not be able to reduce the number of elements much further. In the attached screenshot, a quarter of one ply of your structure is meshed with 750 elements (and you could use a coarser mesh in the theta direction), while using 5 elements through the thickness of the ply, which should be plenty. That's 3,000 elements per ply for the entire structure. I can't imagine that you have 300+ plies.
What am I missing?
Jeff
Hello Dominique, I am not sure how you mesh that hollow half sphere, but I can't see how you would not be able to reduce the number of elements much further. In the attached screenshot, a quarter of one ply of your structure is meshed with 750 elements (and you could use a coarser mesh in the theta direction), while using 5 elements through the thickness of the ply, which should be plenty. That's 3,000 elements per ply for the entire structure. I can't imagine that you have 300+ plies. What am I missing? Jeff


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 28 feb 2011, 10:03 GMT-5
I'm having difficulties to use the Swept mesh since my surface has only 3 edges and not 4. How did you achieve a swept with edges having a mapped mesh without the need of converting it to triangles?

Also, doesn't the top portion of the mesh with very elongated triangles (high aspect ratio and high skewness) could cause problem in the solving process? I've always thought that a good mesh needed to have more or less compact elements?

(Even at 1 million elements, I get a "Out of Memory LU" error...)
I'm having difficulties to use the Swept mesh since my surface has only 3 edges and not 4. How did you achieve a swept with edges having a mapped mesh without the need of converting it to triangles? Also, doesn't the top portion of the mesh with very elongated triangles (high aspect ratio and high skewness) could cause problem in the solving process? I've always thought that a good mesh needed to have more or less compact elements? (Even at 1 million elements, I get a "Out of Memory LU" error...)

Jeff Hiller COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 28 feb 2011, 11:00 GMT-5
Dominique,

The same meshing approach carries over to your hollow half sphere. This is done in the attached file. Notice that I built the geometry by patching together four quarters of the geometry. This introduces internal boundaries that are used for meshing.

Regarding your second question: you can use the Distribution nodes to control the distribution of your mesh; this gives you complete freedom over your mesh distribution. The attached screenshots show this.

Jeff
Dominique, The same meshing approach carries over to your hollow half sphere. This is done in the attached file. Notice that I built the geometry by patching together four quarters of the geometry. This introduces internal boundaries that are used for meshing. Regarding your second question: you can use the Distribution nodes to control the distribution of your mesh; this gives you complete freedom over your mesh distribution. The attached screenshots show this. Jeff


Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 28 feb 2011, 11:42 GMT-5
Thanks Jeff.

The problem I encounter is that I cannot make a circular sweep, I always get the following error:
"Sweep path 'circular' is not supported on this geometry".
(It's able to start a sweep when I use the straight line option but it gives a very bad sweep which does not follow my geometry at all).

I find that weird since my geometry is based on a sphere with layers. Could it be that new "layer" option that is problematic?

Or could it be the Boolean operations I used to create a fourth of an empty sphere?

Here is my model included.

If you could point me out in the right direction it would be really appreciated.

Thanks,
Dominique
Thanks Jeff. The problem I encounter is that I cannot make a circular sweep, I always get the following error: "Sweep path 'circular' is not supported on this geometry". (It's able to start a sweep when I use the straight line option but it gives a very bad sweep which does not follow my geometry at all). I find that weird since my geometry is based on a sphere with layers. Could it be that new "layer" option that is problematic? Or could it be the Boolean operations I used to create a fourth of an empty sphere? Here is my model included. If you could point me out in the right direction it would be really appreciated. Thanks, Dominique


Jeff Hiller COMSOL Employee

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 28 feb 2011, 12:12 GMT-5
Hello Dominique,
Please contact COMSOL's tech support team. We can help you from there.
Jeff
Hello Dominique, Please contact COMSOL's tech support team. We can help you from there. Jeff

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.