Re: [Libmesh-users] how to read a distributed mesh

[Jahrul A. <al...@mu...>]

Hello, Roy and John,

Thank you very much for your feedback. Yes, I am working with a mesh that
is too big. Either a mesh of 1024^3 cells or a mesh of 256^3 cells that is
further refined adaptively to get a local resolution of 1024^3. It can be
opened on a single processor subject to the available memory but I like to
read and write in parallel, having the mesh distributed at each time step.
Even with a smaller mesh, it is question of "algorithmic" view, if I can do
so, and if I can utilize libMesh for this. I fully understand John's work
around.

I agree with Roy that serialized meshes are sometimes useful. However, I
need to deal with a large mesh and associated solution vector, and read the
transient data for post-processing purpose - for which parallel read/write
is useful.

If I understood Roy, I could use either Nemesis format or Checkpoint format
for parallel read/write. Please correct me if this is NOT the case. I will
try these two formats.

Thank you very much for helping computational researchers with the libMesh
package.

Regards,

Jahrul




On Mon, Dec 4, 2017 at 3:40 PM, Roy Stogner <roy...@ic...>
wrote:

>
> On Mon, 4 Dec 2017, John Peterson wrote:
>
> On Mon, Dec 4, 2017 at 10:59 AM, Alam, Jahrul <al...@mu...> wrote:
>>
>> Thanks for the clarification. Can you suggest a work around? If I do,
>>> mesh0.read(lshaped_0.vtu) and mesh1.read(lshaped_1.vtu), and so on, is it
>>> possible to end up with a DistributedMesh by combining mesh0,
>>> mesh1,............. at the user level?
>>>
>>> Or, is there another format that supports parallel read and write?
>>>
>>
>> I wouldn't worry about parallel read/write until you get to a point that
>> you are working with a Mesh that's too big to even be opened on one
>> processor.
>>
>> Your code can still run in parallel even when you read a mesh in serial.
>>
>
> For the sake of completeness, I'll point out that you can read and
> write distributed meshes in either the standard Nemesis format or in
> a libMesh-specific (and libMesh-version-specific! beware!)
> "Checkpoint" format.  (Our libMesh-specific XDR format let you write
> solution data in parallel, but not mesh data currently)
>
> I would definitely suggest taking John's advice, though.  For archival
> purposes of final solutions, serialized mesh files are just easier to
> deal with.  Parallel mesh files become useful when you have so much
> data to write out (transient problem with a good solver with which you
> want to make a movie?) that I/O becomes a bottleneck.
> ---
> Roy
>
>
> ------------------------------------------------------------
> ------------------
> 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
>



-- 
=====================================
Jahrul Alam, PhD
Assistant Professor, Dept of Mathematics and Statistics
Memorial University, St John's, NL, Canada
http://www.math.mun.ca/~alamj/
eMail: al...@mu...
Office: HH-3035, Tel: 709-737 8071
=====================================