HDF5 nuclear data converted from ACE

User Support
1

I am able to convert ACE data to HDF5 (using the provided script) and run OpenMC (develop and various other branches) with that data locally. I am also able to run OpenMC with that locally-converted data on a cluster. However, when I do the conversion on the cluster, running OpenMC with the converted HDF5 data yields:

HDF5-DIAG: Error detected in HDF5 (1.8.18) MPI-process 0:
#000: H5F.c line 604 in H5Fopen(): unable to open file
major: File accessibilty
minor: Unable to open file

I don’t think it’s an issue of the files’ existence or permissions. Are there any typical causes of this error when running OpenMC?

Thanks,
Jon

2

Hi Jon,

Is it possible that the filesystem the data is a local filesystem on a single node (i.e., other ranks don’t have access to the file)?

Best,
Paul

3

With all data on the same filesystem, I am able to access HDF5 files that were converted from ACE locally and then transferred to the cluster, but when I hit a file (in the same run) that was converted on the cluster, I hit this error.

I have ~TBs of ACE data I’m trying to convert, so I’d prefer to keep it on the cluster.

4

What version of HDF5 is your h5py installation linked against? You can find out by running

import h5py
print(h5py.h5.get_libversion())

5

In the meantime, I built a new HDF5, and with gcc-4.9.3, mvapich2-2.2, and hdf5-1.10.1 everything is now working on the cluster.

That said:

– The erroring exe was built with gcc-4.9.3, mvapich2-2.2, and hdf5-1.8.18 with the same configuration as the successful hdf5-1.10.1

– cluster h5py is linked against 1.10.0

– local h5py is linked against 1.8.13

So, the 1.8.18 openmc didn’t play well with the 1.10.0-generated data, but the 1.8.13-generated data was fine (as is 1.10.0-generated data with a 1.10.1 exe).

Thanks for looking at this, Paul.

6

This is probably the line that bit you – we use libver=‘latest’ when creating HDF5 files using h5py, which may produce libraries that are incompatible with previous versions of HDF5 as described here. Glad you got it working!