PMPI Error

User discussions
1

GROMACS version: gmx grompp, version 2018
GROMACS modification: No
I was attempting to simulate 3 transmembrane proteins in a lipid bilayer with Martini representation on Bridges when I received the following error:

PMPI_Sendrecv(236): MPI_Sendrecv(sbuf=0x7fffb52e9880, scount=8, MPI_BYTE, dest=20, stag=0, rbuf=0x7fffb52e9980, rcount=8, MPI_BYTE, src=80, rtag=0, MPI_COMM_WORLD, status=0x7fffb52e9750) failed

The PSC Bridges consultants were unfamiliar with this error, though I successfully ran a nearly identical system with the proteins more tightly clustered without this error. I’m happy to provide additional details, and thank you in advance for your time.

2

Is this part of a larger backtrace from a segmentation fault? Is there any GROMACS-specific error? The system could just be unstable and is blowing up.

3

Thanks for your help. The preceding lines are:
Steepest Descents:
Tolerance (Fmax) = 1.00000e+01
Number of steps = 10000
Abort(402774798) on node 8 (rank 8 in comm 0): Fatal error in PMPI_Sendrecv: Message truncated, error stack:
PMPI_Sendrecv(236): MPI_Sendrecv(sbuf=0x7fffb52e9880, scount=8, MPI_BYTE, dest=20, stag=0, rbuf=0x7fffb52e9980, rcount=8, MPI_BYTE, src=80, rtag=0, MPI_COMM_WORLD, status=0x7fffb52e9750) failed
(unknown)(): Message truncated

4

So the fault is not with grompp but mdrun. What is your system and how did you construct it? If you’re getting an immediate failure, either the MPI implementation is buggy (so test by using OpenMP or a single thread) or the system is entirely unstable and causes mdrun to immediately choke.

5

The system is 3 membrane-integrated proteins, where the crystal structures were passed from RCSB to charmm-gui to build the MARTINI representation membrane system. Checking the log file, I see that 24/10,000 of the minimization steps are taken before the crash. I will test with a single thread to test the mpi implementation.