ParaView cannot open BP files

Hi,

I am really excited about this module. We have a number of really large cases that could benefit from using more clever ways of checkpointing or writing output and this looks like a step in the right direction.

I built it with version 2006 and run some tutorials. BP files were generated, but when I attempt to open those in ParaView it breaks with the following error:

free(): invalid size

Loguru caught a signal: SIGABRT
munmap_chunk(): invalid pointer

Loguru caught a signal: SIGABRT

Loguru caught a signal: SIGSEGV

I am using ParaView downloaded directly from Kitware website rather than built against OpenFOAM distribution? I guess this should not be important given ADIOS is adaptable, but please let me know.

Probably a bit of misunderstanding here. You can think of ADIOS like a data container that specifies a data format, but not its content, maybe like a tar or zip container. A java jar, office docx, pptx are examples of different things using zip containers, but with radically different content.

Thanks for the reply, Mark. Yes, I knew about ADIOS being a container, but didn't connect the dots! I have now realised that adios comes with bpls which lets you view this just like h5ls in HDF5.

I will close this, but I will take this opportunity to ask a few brief questions:

1. Can this module be used for checkpointing? Yes. You state this in the introduction.
2. Is the programming interface the only way to us the data stored in the container?
3. Have you observed benefit in I/O interactions on cases larger than 1000+ subdomains?

EDIT: answered the first question.

Hi @robertsawko - got caught doing something else yesterday.

At the moment, the data are indeed stored opaquely with the programming interfacing currently being the only direct access. In the future, it would be obviously be desirable to make this storage mechanism a first-class object. However, if you follow the trail of OpenFOAM internal with the various IOobject READ_IF_MODIFIED etc, it will quickly become apparent that replacing/augmenting the OpenFOAM stream-based IO will not be easy.

The IO benefits for cases larger than 1k processes might be something that @ivanspisso or @williamfgc might have information on.

Good Morning, we are still "blocked" on such activity with @williamfgc We should think a work-plan for a big I/O test-case on Tier-0 machines...

Thanks to you both. I will close this, but I hope you don't mind if I follow-up by email at some point.

We have some problems with checkpointing for large cases (cells O(100m)-O(1B) of cells and 1k< subdomains). Both reliability and speed are an issue. I do have access to some large systems with GPFS and would be happy to get involved in some amount of benchmarking as it could our UK team.

We also store lots of other objects such as 3D contours and slices in postProcessing. We then write bespoke VTK application that process those further, but I need to think whether ADIOS offers any benefit other than tidying it all into one file.

closed

Dear @robertsawko I am more that happy to disscuss it inside to the HPC TC https://wiki.openfoam.com/High_Performance_Computing_(HPC)_Technical_Committee Please, continue discussion via e-mail: to write to all the partner openfoam-hpc-tc@googlegroups.com, else if you prefer, please write me to i.spisso@cineca.it Regards Ivan