A CFX simulation using the discrete transfer radiation model seems to fail randomly and inexplicably. What are the most likely causes and how can this problem be dealt with?
This problem typically manifests itself as a random, inexplicable failure of the solver when using the discrete transfer radiation model, with the ray tracing files stored on a networked disk. The failure is unlikely to be reproducible, hence a repeat simulation will not fail at the same point. Error messages similar to those shown below may be written to standard output:
PGFIO/stdio: Input/output error
PGFIO-F-/unformatted write/unit=99/error code returned by host stdio - 5.
unformatted, sequential access record = 1725
For a parallel run these messages will be written to the /tmp/pvml file on the host concerned, although they should also be echoed to the solver output file. Additionally, error messages similar to that shown below may also be generated:
ERROR #333000014 has occurred in subroutine RCVBUF.
Problems receiving a PVM-buffer from part. 3 to part. 1
--> PVM-error flag from notifying missing partition: -31
The problem is that thecode used to access the ray tracing files can be susceptible to network instabilities when these files reside on networked disks. Beware of using NFS scratch servers to store these files. The solution is to use a local disk, such as /tmp on UNIX systems. Quite apart from this, since the ray tracing files are heavily accessed during the radiation solution, localised storage can significantly improve performance. The ray tracing file path is specified as follows:
'Solver Control > Advanced > Thermal Radiation Control > Ray Tracing Control > File Path'
Before starting a simulation with a specified ray tracing file path, always make sure that the location (on all hosts, if distributed parallel) is free of any extraneous ray tracing files, or the new run will write the ray tracing files to the default location (working directory) for the run. If a run terminates normally, any ray tracing files should be removed automatically, but are likely to be present after a failed run. Ray tracing files are named rtrace.rg* Finally, ray tracing files can be large, often several hundred MB in size, hence the available disk space should always be checked.