MPI Installation and Configuration
- Have a look to the section Repository Settings and ensure that you have the common repo files
- Before starting the installation procedure remember to clean all yum cache and headers:
yum clean all
MPI Service installation on CE-CREAM
Follow the CE-CREAM installation guide before starting to follow this steps below
yum install mpi-start glite-yaim-mpi
MPI Service installation on WN
Follow the WN installation guide before starting to follow this steps below
yum install mpi-start glite-yaim-mpi
yum install openmpi openmpi-devel mpich2
- Have a look to the MPI documentation before starting to install: MPI Guides
.
IMPORTANT NOTE: |
Remember if you are using Torque to create the munge key, and copy it to all cluster hosts (CE, Batch Master, WNs): Munge configuration |
Service Configuration
Useful Variables
Remember to copy these three files in your service directory:
- glite-mpi
- glite-mpi_ce
- glite-mpi_wn
Please set properly the variables in the files in service directory. In particular customize this important values:
File name |
Variable |
common value |
Description |
glite-mpi |
MPI_MPICH_ENABLE |
MPI_MPICH_ENABLE = "no" |
Support for MPICH Flavour |
glite-mpi |
MPI_MPICH2_ENABLE |
MPI_MPICH2_ENABLE="yes" |
Support for MPICH2 Flavour |
glite-mpi |
MPI_OPENMPI_ENABLE |
MPI_OPENMPI_ENABLE="yes" |
Support for OPENMPI Flavour |
glite-mpi |
MPI_MPICH2_PATH |
MPI_MPICH2_PATH="/usr/lib64" |
MPICH2 path |
glite-mpi |
MPI_MPICH2_VERSION |
MPI_MPICH2_VERSION="1.2.1p1" |
MPICH2 version |
glite-mpi |
MPI_OPENMPI_PATH |
MPI_OPENMPI_PATH="/usr/lib64/openmpi/1.4-gcc/" |
OPENMPI path |
glite-mpi |
MPI_OPENMPI_VERSION |
MPI_OPENMPI_VERSION="1.4-4" |
OPENMPI version |
glite-mpi |
MPI_MPICH_MPIEXEC |
MPI_MPICH_MPIEXEC="/usr/bin/mpiexec" |
MPICH MPIEXEC path |
glite-mpi |
MPI_MPICH2_MPIEXEC |
MPI_MPICH2_MPIEXEC="/usr/bin/mpiexec" |
MPICH2 MPIEXEC path |
glite-mpi |
MPI_OPENMPI_MPIEXEC |
MPI_OPENMPI_MPIEXEC="/usr/lib64/openmpi/1.4-gcc/bin/mpiexec" |
OPENMPI MPIEXEC path |
glite-mpi |
MPI_SSH_HOST_BASED_AUTH |
MPI_SSH_HOST_BASED_AUTH=${MPI_SSH_HOST_BASED_AUTH:-"yes"} |
Use the SSH Hostbased Authentication between your WNs |
glite-mpi_ce |
MPI_SUBMIT_FILTER |
MPI_SUBMIT_FILTER=${MPI_SUBMIT_FILTER:-"yes"} |
For Torque ensure that CPU allocation is performed correctly |
YAIM Verification
- Before starting the configuration PLEASE TEST that you have defined all the mandatory variables for the MPI profile
-
- On CE-CREAM MPI Verification
/opt/glite/yaim/bin/yaim -v -s <site-info.def> -n MPI_CE -n creamCE -n TORQUE_server -n TORQUE_utils
/opt/glite/yaim/bin/yaim -v -s <site-info.def> -n MPI_WN -n WN_torque_noafs
You can find in this documentation: YAIM MPI Variables all mandatory variables.
If no errors are reported during the verification you can proceed to the configuration, otherwise correct them before continuing.
YAIM Configuration
- Please use the debug flag (
"-d 6" ) to configure the services in order to have detailed information.
- For your convenience yo can save all the configuration information in a log file for each configuration separately, different from the default
yaimlog , that contins all the history of all configurations.
IMPORTANT NOTE: |
Remember to put at first position the nodetype MPI_CE or MPI_WN |
/opt/glite/yaim/bin/yaim -c -d 6 -s <site-info.def> -n MPI_CE -n creamCE -n TORQUE_server -n TORQUE_utils 2>&1 | tee /root/conf_EMI_CREAM_Torque_MPI.`hostname -s`.`date +%Y-%m-%d-%H-%M-%S`.log
/opt/glite/yaim/bin/yaim -c -d 6 -s <site-info.def> -n MPI_WN -n WN_torque_noafs 2>&1 | tee /root/conf_WN_Torque_MPI.`hostname -s`.`date +%Y-%m-%d-%H-%M-%S`.log
Check files in WN
If in YAIM Configuration you are chose to use MPI_SSH_HOST_BASED_AUTH (well-advised) check these files:
- /etc/ssh/sshd_config
- /etc/ssh/shosts.equiv
The first one sshd_config should have the follow variables set like below
HostbasedAuthentication yes
IgnoreUserKnownHosts yes
IgnoreRhosts yes
The second file shosts.equiv should contains: the CE hostname, the default SE host name and all the WNs hostname. May be in the CE it was created in configuration process, copy it form the CE.
Restart sshd services after file modification:
service sshd restart
Service Testing - Reference Card
After service installation to have a look if all were installed in a proper way, you could have a look to User & Service Troubleshooting Guide . In this page you can find some common errors and other useful service information.
MPI Documentation References:
|