nrniv -mpi Error

[calvins]

I downloaded and installed NEURON 6.0 following Hines, M.L. and Carnevale, N.T. "Translating network models to parallel hardware in NEURON" on a Fedora Core release 3 unit with 2 processors (with the --with-paranrn configure option, etc). I received the correct output for the MPICH2 cpi example, shown below

mpiexec -l -n 2 ./cpi
0: Process 0 of 2 is on server
1: Process 1 of 2 is on server
0: pi is approximately 3.1415926544231318, Error is 0.0000000008333387
0: wall clock time = 0.000621

When I run the test0.hoc without the -mpi, I get the following:

mpiexec -l -n 2 nrniv -nobanner -nogui test0.hoc
0: There are 1 processes. My rank is 0 and I am on server
1: There are 1 processes. My rank is 0 and I am on server

When I add the -mpi command, I get the following:

mpiexec -l -n 2 nrniv -nobanner -nogui -mpi test0.hoc
0: MPI process rank 0 (n0, p19736) caught a SIGSEGV in MPI_Comm_dup.
1: MPI process rank 0 (n0, p19735) caught a SIGSEGV in MPI_Comm_dup.
0: Rank (0, MPI_COMM_WORLD): Call stack within LAM:
0: Rank (0, MPI_COMM_WORLD): - MPI_Comm_dup()
0: Rank (0, MPI_COMM_WORLD): - main()
1: Rank (0, MPI_COMM_WORLD): Call stack within LAM:
1: Rank (0, MPI_COMM_WORLD): - MPI_Comm_dup()
1: Rank (0, MPI_COMM_WORLD): - main()

Have you seen this error before or know how to fix it?

[ted]

What happens if you use the most recent standard distribution of NEURON?

[calvins]

Hi Ted,
With NEURON 7.0, I get the same error. I'm trying to get parallel NEURON to work on the two-processor computer before I try to set it up in a cluster. Any suggestions for what I could do to try to fix the error? Thanks.

[hines]

correct output for the MPICH2 cpi example

mpiexec -l -n 2 nrniv -nobanner -nogui -mpi test0.hoc
...
0: Rank (0, MPI_COMM_WORLD): Call stack within LAM:

There seems to be a mixture of mpich2 and lam on your machine.
NEURON configure looks for mpic++ before it looks for mpicxx.
What do you get for
which mpicc
which mpic++
which mpicxx
are they all coming from the same place. If not go to the mpich2 bin directory and
cp mpicxx mpic++
and try doing a clean build again and make sure the mpich2 bin directory is first in your path so
mpicc and mpic++ are coming from the same place.

[calvins]

You were 100% correct. The mpic++ file was not found in the mpich2 directory and copying the mpicxx and reinstalling worked. You guys are amazing, thanks!