H5Part development and discussion

["Adelmann Andreas (PSI)" <andreas.adelmann AT psi.ch>]

Dear Jozef Achim Gsell the main developer of H5Part (we call it now H5hut) and can definitively

explain you more in detail. We have in H5hut the capability to store block structured fields (scalar and vector) 

that is working for your points 1 … 3, not sure about 4. We are also working on and off on unstructured grids. The latter does not have priority on our side but (==off)

but definitely you are very welcome to join forces. 

Sure Achim can give more details. Maybe we should have a Skype (or hangouts) meeting to discuss details.

Cheers Andreas 

------
Dr. sc. math. Andreas (Andy) Adelmann
Staff Scientist
Paul Scherrer Institut WBBA/219 CH-5232 Villigen PSI
Phone Office: xx41 56 310 42 33 Fax: xx41 56 310 31 91
Phone Home: xx41 62 891 91 44
-------------------------------------------------------
Friday: ETH CAB F37  +41 44 632 75 22
============================================
The more exotic, the more abstract the knowledge, 
the more profound will be its consequences.
Leon Lederman 
============================================

On 24 Oct 2016, at 18:22, Jozsef Bakosi <jbakosi AT lanl.gov> wrote:

Hi folks,

I realize H5Part is for particles, but I wonder how much effort the following
would take.

I have been looking for a library that is capable of reading and writing large
data sets (1) in a portable fashion, e.g., using NetCDF and/or HDF5, (2) in
parallel, and (3) preferably to and from a single file. Geometrically, the data
sets to be stored are defined at nodes of a tetrahedra-only unstructured mesh +
the mesh itself (node coordinates and element connectivity). Additionally, (4)
the mesh (number of nodes + connectivity) may change over different time steps.

I have been using ExodusII (https://github.com/gsjaardema/seacas) for this
purpose, which works okay, satisfying (1) and (2), but not (3), and (4) can only
be fulfilled with separate files per time step. Considering that ExodusII saves
a single file per parallel rank already, having to save a new file per time step
will not scale for large data and many processors.

I'm also using H5Part to save particles and now I'm thinking that what H5Part
does already might be pretty close to what I need: it can already save a set of
point (particle) coordinates + associated field data whose numbers can change
over time steps. This satisfies (1), (2), and (3) while (4) could be satisfied
by adding the capability to save the mesh connectivity, a set of integers
connecting the point IDs (while the points would be saved similarly to
particles).

My questions are:

1. Does anyone know of an existing library that satisfies all the above
requirements?

2. How much work do you think it would be to add the mesh topology information
to H5Part? If it isn't too bad, I would like to contribute this so H5Part can be
used to save meshes, field data associated to mesh nodes, as well as the
existing particles and field data associated to particles. I would also extend
the ParaView H5Part reader plugin.

Any thoughts and advice are appreciated.

Thanks,
Jozsef
--
Jozsef Bakosi
Computational Physics and Algorithms (CCS-2)
MS D413, Los Alamos National Laboratory
Los Alamos, NM 87545, USA
email: jbakosi AT lanl.gov
phone: 505-665-0950
fax: 505-665-4972