Discussion Forum

August 26, 2009 6:52pm UTC

Out of memory during LU factorization

Hi there,

I am a new member in the COMSOL community. I am trying to solve a simple 3D problem having diffusion and chemical reaction within the subdomain. However, I am getting this message "out of memory during LU factorization." Is my computer running out of memory to solve this particular problem or should I use a better solver to solve this problem? Could anyone share his/her experiences with COMSOL?

Many thanks,

Mohammed

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse

August 26, 2009 10:36pm UTC in response to F Toor

Re: Out of memory during LU factorization

Hi Fatima,

Many thanks. I tried both the solvers (UMFPACK and SPOOLES), but still couldn't resolve the problem. How about using different mesh elements?

Mohammed

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse

August 27, 2009 8:05am UTC in response to Mohammed Hussain

Re: Out of memory during LU factorization

Hi

there is another way to get a little more RAM, works successfully for me, by using the server-client approach as then the client can be swapped during the solve processing, this has allowed me to rougly double the DoF I can solve on my laptop (when I was using a laptop, now I'm on a larger server with much more RAM and havnt used this approach for some time).

This does not mean that you should reduce the mesh size to get your model smaller, with the risk o less accuracy

Playing with the sovers also can help, but that again depends largely on the problem you are solving, symmetric or no, linear or not ... , there arequite some developments about this in the doc, sometimes with "local" application mode tricks in the specific application mode documentation, generally under the title "solver"

Good luck
Ivar

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse

August 31, 2009 8:28pm UTC in response to Joshua Golbert

Re: Out of memory during LU factorization

Hi

What you mean, I beleive is that with more addressable RAM, as you can use on a 64bit (addressable) CPU, you can digest lager models.
That is ceranly true, just as well as you can run multiple processors in parallel, too, and substantially speed things up.
Unfortunately, you will not gain in precision on the numerics, as the "64 bits" do normally not imply any change to the floating point processor.

But, I have heard that GPU parallel processors might be soon used too, so if you are really looking for large models, they will soon be as "easy" to solve as simple 2D sketches are today

Ivar

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse

November 26, 2009 12:56pm UTC in response to Eyal Cohen-Hoshen

Re: Out of memory during LU factorization

Hi,

recently i have done some quite strange experience with this. When i run simulation on WXP, i encounter "out of memmory" error, but when i run same simulation on the same machine using ubuntu 9.04, the computer manages to acomodate the data into RAM and, according to the sounds which come out of the machine(HDD runs for its life), also the in SWAP partition. Not suprisingly, the simulation runs slow because of the hard drive data flow limit. Ergo it seems, that WXP somehow cant use the paging file. I have tried increasing the size of it to 8 GB with no effect. During the simulation, that crashes because of lack of memory, the page file grows about 100MB and not more even though the RAM usage grows over 1G and than the "out of memory" error comes, so the use of paging file would be expected. Another interesting thing is, that when I run simulation on WXP, which is small enough to fit in RAM only and therefore it succesfully finishes, it runs significantly faster, than on ubuntu 9.04 ( approx 40% faster - same task, same solver, etc. ). Some interesting thing is, that when you look on the CPU usage graphs during calculatin in both systems, you can see why the computing time is so signficantly longer. In WXP both my cores (i have dual core 32 bit processor) have almost same performance graphs during the evaluation. But in UBUNTU, the cores never work on 100% at one time. The 100% performance jumps from one core to another in about five seconds(approx, irregularly). The second core is never used more than 30%.

So long story short : during my calculation in WXP the system fails to use paging file(and generates "out of memmory" error), but in ubuntu 9.04, the system fails to split the calculation effectively for the CPU cores and runs significantly slower.

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse

May 19, 2010 9:33am UTC in response to Mohammed Hussain

Re: Out of memory during LU factorization

Dear Mohammad,

could you solve your problem? I encounter out of memory and out of memory LU decomposition errors. I don't know how to deal with these memory errors. please guide me.


va saalamo alaikoom va rahmatollah
With Best regards,
Fatemeh Sharifi

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse

October 2, 2010 2:56pm UTC in response to Ivar Kjelberg

Re: Out of memory during LU factorization

how can i active server-client approuch?

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse

December 25, 2011 8:27am UTC in response to Ivar Kjelberg

Re: Out of memory during LU factorization

I solved once a laser welding model. Resolving again of that model encountered to this error.
No changing the mesh or solver helped...

my system: quad4, 4gb ram, windows 64 7...comsol 4.2

and anybody know why UMFPACK is absent in comsol 4.2??

Thanks for help...

Reply  |  Reply with Quote  |  Send private message  |  Report Abuse