Problems with large CGNS files of unstructured arbitrary polyhedral data

We’re encountering significant challenges when working with large unstructured CGNS files, specifically those with arbitrary polyhedral mesh and solution data. Our files often contain upwards of 350 million cells, and we’re frequently running out of memory during processing. This is on systems with upwards of 300GB memory.

We’ve attempted to utilize the distributed memory parallelism across several nodes to manage these large files. However, we’ve observed through node monitoring that only the master process is engaged in reading the files, which seems to limit the effectiveness of the approach.

Could anyone provide insights or advice on the following:

  1. Are we possibly misconfiguring our setup that causes only the master processor to handle the file reading? If so, any guidance on proper configuration for efficient distributed reading would be highly appreciated.
  2. If our setup is correct, what alternative strategies or tools could we consider to handle large CGNS files more effectively without running into memory constraints?

Any advice or experience with similar issues would be greatly appreciated!

Thank you in advance!

Please share a typical yet small datafile.