TURBOMOLE Users Forum
Installation and usage of TURBOMOLE => Parallel Runs => Topic started by: cavallo on June 19, 2008, 02:03:02 PM
-
Hi,
trying to run dscf on an opteron dual core. This is the script I use is:
#!/bin/sh
PARNODES=4
/home/programs/TM-5.10/TURBOMOLE/bin/x86_64-unknown-linux-gnu_mpi/dscf > dscf.log
plus I put $numprocs 4 in the control file. The MACHINE FILE simpy contains 4 line with the machine name. I have 4 slave outputs. So, everything seems normal.
What is strange, is that after 10 minutes I see 5 dscf jobs using cpu. I expected 4 using cpu, the fifth just spawning processes, so that the used cpu should be almost 0.
PID TIME+ COMMAND
13254 10:49.12 dscf_mpi
13255 11:22.49 dscf_mpi
13258 11:18.50 dscf_mpi
13257 11:25.62 dscf_mpi
13256 11:01.32 dscf_mpi
The hp_mpi_appfile created by dscf contains 5 lines as the one below:
-h k116 -e LD_LIBRARY_PATH=/home/programs/TM-5.10/TURBOMOLE/libso/em64t-unknown-linux-gnu_mpi:/home/programs/ifort/lib:/home/programs/TM-5.10/TURBOMOLE/libso/x86_64-unknown-linux-gnu_mpi/ -e MPI_REMSH=/usr/bin/ssh -e MPI_WORKDIR=/scratch/cavallo/n -e MPI_FLAGS=y0 -np 1 /home/programs/TM-5.10/TURBOMOLE/bin/x86_64-unknown-linux-gnu_mpi/dscf_mpi
Is this normal ? Why all 5 dscf_mpi are sucking cpu time ?
TIA,
Luigi
-
I just found the "TM 591 parallel - to many computing(!) processes" thread, and I discovered that adding -intra=nic to the mpirun execution line in dscf solved the issue.
Sorry for duplicating a post.
Luigi