[me3840]

Hello,

I have a cluster I'm compiling OpenFOAM on. I have tried both the foam-extend 4.0 and OpenFOAM 5.0 which have the same issue, so I suspect I'm doing something wrong.

The compilation of both is successful, however when I try to read any binary file, I get errors similar to:

Expected a '}' while reading binaryBlock...

if I write meshes with ASCII they will read and decompose just fine.

Since my compilation on my workstation (Centos 5) works just fine, and the cluster is based on Scientific Linux (which is pretty similar to Centos 5) I imagine most of what I do should be pretty similar. No errors get thrown during compilation after I load all the needed modules.

How can I test what could be the issue here?

Thanks

[wyldckat]

Greetings me3840,

That's a pretty bizarre error message, within the context of what you described. My guesses would be:

1. That the header files being used during compilation, do not match the libraries being used at runtime.
   • For example, this could make sense if the cluster is using Big Endian byte order for binary data, while your workstation is using Little Endian: https://en.wikipedia.org/wiki/Endianness - and this is assuming you were then using the simulation case on both machines, e.g.: decomposing on the workstation and then simulating on the cluster.
   • The same could happen, if the case is decomposed on the login node of the cluster, but then simulated on the cluster nodes, in case they use different Endianness.
2. There could be some library contamination when launching the case on the cluster nodes, for example, the shell environment on the login node does not match the same environment on the cluster nodes.

To figure out what exactly is going on, it's fairly simple:

1. Copy the tutorial case "$FOAM_TUTORIALS/incompressible/icoFoam/cavity/cavity". This case is so simple, that the results should be pretty much identical on any machine.
2. Configure the file "system/controlDict" to write as binary.
3. Run the case in serial mode on the following machines:
   1. The workstation.
   2. The login node on the cluster.
   3. On a normal cluster node (via job scheduler?).
4. Then take the 3 cases, place them in your workstation and use an hexadecimal editor to look into the file "0.5/p" on each case. With luck, you can pinpoint what bytes are different and how exactly different they are.

Beyond this, it would be good to know which compiler name and version you used and how you compiled on each machine. For example, if you compiled on the workstation and then copied to the cluster, or vice-versa; or if you simply used the same installation instructions on both machines.
And it would be nice to know which installation instructions you've followed

Best regards,
Bruno

PS: Caution: RHEL and CentOS 5 are no longer maintained, so use it at your own risk