[petsc-dev] Do you have any suggestion in order to reduce my memory footprint?

[Jed Brown]

"Alletto, John M" <john.m.alletto at lmco.com> writes:

> All,
>
> I have a 64 bit linux machine with 256 GByte of memory and 30 processors.
> I am solving Laplace equations and the current solver I am using is taking every bit of memory I have available.

What is the current problem size that's filling your memory?  How
regular is the domain and coefficients?  Barry's suggestion won't
necessarily reduce memory usage, but it does tend to be fast and easy.

How much memory is used by the solver versus other parts of your
application?  Is the current configuration realistic?  It's not worth
minimizing solver memory usage if it's going to be a small part of the
total application, for example.  If you have simple geometry, smooth
coefficients, and a standard discretization, a matrix-free full
multigrid can solve the problem using about 50 bytes of memory per DOF
at a cost of around 5 times as much as a residual evaluation.  But many
applications don't benefit from such a refined solver.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20141019/4569bd9c/attachment.sig>