Re: [Libmesh-users] 答复: SCM causes out of memory

[David K. <dav...@ak...>]

On Sat, Feb 10, 2018 at 10:03 AM, <cau...@gm...> wrote:

> (Sorry for accidentally clicking "send" without the email being finished)
>
> Hi all,
>
>
>
> I am trying implementing the RB method on a model with 44092 DoFs.
> However, I run into the error message below:
>
>
>
> [2]PETSC ERROR: --------------------- Error Message
> --------------------------------------------------------------
>
> [2]PETSC ERROR: Out of memory. This could be due to allocating
>
> [2]PETSC ERROR: too large an object or bleeding by not properly
>
> [2]PETSC ERROR: destroying unneeded objects.
>
> [2]PETSC ERROR: Memory allocated 0 Memory used by process 2545381376
>  (2.37GB)
>
> [2]PETSC ERROR: Try running with -malloc_dump or -malloc_log for info.
>
> [2]PETSC ERROR: Memory requested 1193845248 (1.1GB)
>
>
>
> The code is run with 8 processors. The computer has in total 19GB memory
> and 12 CPUs. I think that the error occurs when the SCM is performed since
> it runs well when the lower bound is set to be 1. It seems that the
> procedure is enormously memory-consuming and my computer cannot afford to
> run it. I wonder if there is a way in which less memory is demanded. Your
> suggestion would be appreciated.
>
>

Are you using -eps_type lapack? That uses a huge amount of memory because
it treats the matrix as dense, so I guess that's why you ran out of memory.
Using lapack is very robust, i.e. it generally converges well, but it is
only appropriate for very small problems. To do larger problems you'll need
to use SLEPc's sparse eigensolvers. Unfortunately you may run into
convergence problems with those for the SCM eigenvalue problems.
Unfortunately there is no easy fix for those convergence issues for SCM as
far as I know, you'd have to read the SLEPc manual and see if there are any
options you can modify to get better convergence.

David





>
> *发件人**:* David Knezevic [mailto:dav...@ak...]
> *发送时间:* 2018年2月10日 22:25
> *收件人:* cau...@gm...
> *抄送:* lib...@li...
> *主题:* Re: [Libmesh-users] SCM causes out of memory
>
>
>
> Hi,
>
>
>
> Looks like you sent an empty email, but I guess from the subject line that
> you're having issues with the SCM? I'm not sure I can offer much help since
> I haven't used the SCM code for a while, but if you can provide some
> details about your problem maybe I can help a bit.
>
>
> David
>
>
>
>
>
> On Fri, Feb 9, 2018 at 8:55 PM, cau...@gm... <
> cau...@gm...> wrote:
>
>
> ------------------------------------------------------------
> ------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> Libmesh-users mailing list
> Lib...@li...
> https://lists.sourceforge.net/lists/listinfo/libmesh-users
>
>
>
>
>