Just a quick note here... I've successfully used SerialMesh on over 10,000
processors... and ParallelMesh on over 100,000

That "128 processors" quote from the paper is quaint ;-)

Derek


On Mon, Jul 14, 2014 at 1:30 PM, John Peterson <[email protected]> wrote:

> On Mon, Jul 14, 2014 at 11:32 AM, Hodge, Neil E. <[email protected]> wrote:
> > I recently sat down and read
> >
> > Kirk, "libMesh: a C++ library for adaptive mesh refinement/coarsening
> > simulations", Engineering with Computers, 2006.
> >
> > In that paper, I found the following:
> >
> > "On distributed memory machines, such as PC clusters, a complete copy of
> > the mesh is maintained independently on each processor. This design
> > decision limits practical 3D applications to on the order of 128
> > processors because of the overhead associated with storing the global
> > mesh."
>
> Wow, I had forgotten the paper said this!
>
> It's definitely not true any more, and the sentence leaves out
> something important, which is that the amount serial mesh "limits" you
> of course depends on the size of the Mesh.
>
> In practice I use SerialMesh for nearly all of my applications, and
> reserve ParallelMesh only for meshes which can't even be opened on a
> single node because they don't fit in the available system memory.  In
> such cases, you have to use the Nemesis format and "split" your mesh
> in a preprocessing step before running your application.
>
> One of the nastier things that we've discovered within the last year
> or two is that std::map (which ParallelMesh uses to store its nodes
> and elements) imposes an incredible memory overhead relative to
> std::vectors, so that for small enough problems, running with
> ParallelMesh actually requires *more* memory than running with
> SerialMesh, even with the saving you get from parallelizing it
> factored in.
>
>
> >  Is it possible to separate these two functionalities, i.e., to use the
> mesh manipulation functionality, but not perform the element operations?
>
> Absolutely.
>
> --
> John
>
>
> ------------------------------------------------------------------------------
> Want fast and easy access to all the code in your enterprise? Index and
> search up to 200,000 lines of code with a free copy of Black Duck&#174;
> Code Sight&#153; - the same software that powers the world's largest code
> search on Ohloh, the Black Duck Open Hub! Try it now.
> http://p.sf.net/sfu/bds
> _______________________________________________
> Libmesh-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/libmesh-users
>
------------------------------------------------------------------------------
Want fast and easy access to all the code in your enterprise? Index and
search up to 200,000 lines of code with a free copy of Black Duck
Code Sight - the same software that powers the world's largest code
search on Ohloh, the Black Duck Open Hub! Try it now.
http://p.sf.net/sfu/bds
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to