Internal SciDB error, gesvd/mpi


#1

Folks,
Any thought on the error below from SciDB 13.3. MPI version is MPICH2.

iquery -aq "gesvd(productH,‘U’)"
SystemException in file: src/mpi/MPIManager.cpp function: initMpiLinks line: 135
Error id: scidb::SCIDB_SE_INTERNAL::SCIDB_LE_SYSCALL_ERROR
Error description: Internal SciDB error. Invocation of symlink failed with return code=-1, error=2, arg(s)=/bigdata/scidb/merra/000/0/mpi.
Failed query id: 1100889585652

FYI: I have no trouble running non-SciDB MPI applications on the system.

Thanks,
-Hamid


#2

Hamid -

Sorry about the delay in this response. Some thoughts:

  1. MPI setup is subtle and tends to be fragile. I don’t want this to come across as an RTFM answer, but we spent a lot of time/effort documenting the procedure in the user’s guide. I would advice you first to double check that you followed the procedure described there.

  2. The message is pretty clear, but it’s meaning is a bit opaque. To help us, would you please provide:

  • The operating system version.
  • The version of MPICH2 (that’s the name of the package - it has it’s own version number).
  • Have you tried this again on our 13.6?

Please let us know how you go. I suspect (as you mention that you have no problem running non-SciDB MPI programs) that there’s a version mis-match between what SciDB expects and what you have installed. We pay a lot of attention to testing our dependencies which means SciDB has adopted the practice of fixing versions in the software packaged we depend on. Consequently, we might not work with the “latest” release of something.


#3

Plumber,
Thank you very much for your response.
I will check the user’s guide again to ensure I have not overlooked something. BTW, which section of the user’s guide discusses MPI configuration? The only references to MPI I see in 13.3 user’s guide is Appendix A.4 pp 240 (MPICH2 license), Appendix B pp 242 (Acknowledgements) and the reference to ScaLAPACK/MPI on page 7.

Answers to your questions:
OS: CentOS release 6.3 (Final)
MPICH2: 1.2.1
We are in the process of upgrading to 13.6. I will try things out once that is done and let you know.

-Hamid