Appendix E: Using NVIDIA SHARP from DOCA-Host
When using DOCA-Host distribution, the HPCX_SHARP_DIR environment variable has to be set to redirect to the SHARP installation directory (default location: /
opt
/mellanox/sharp), and OMPI_HOME environment variable to the
MPI
installation directory.
To download the DOCA-Host packages, please visit here.
SHARP_am can be run from UFM or as part of the installation of DOCA-Host. When installing from DOCA-Host, make sure SHARP_am is being installed on the system as daemon.
The minimal required OpenSM version for working with SHARP_am is v5.20.
This section describes how to install Aggregation Manager in the fabric using NVIDIA SHARP_am daemon script.
NVIDIA SHARP Aggregation Manager daemon (sharp_am) is executed on a dedicated server along with the Subnet Manager.
Installing Aggregation Manager as a service is required when used from the HPC-X or DOCA-Host packages.
NVIDIA SHARP Installation Script
In order to install or remove NVIDIA SHARP_am daemons, use the sharp_daemons_setup.sh script provided with the NVIDIA SHARP package. For example:
$HPCX_SHARP_DIR/sbin/sharp_daemons_setup.sh
Usage: sharp_daemons_setup.sh (-s | -r) [-p SHARP location dir] -d
<sharpd | sharp_am> [-m]
-s - Setup SHARP daemon
-r - Remove SHARP daemon
-p - Path to alternative SHARP location dir
-d - Daemon name (sharp_am)
-b - Enable socket based activation of the service
Registering SHARP_am as a Service on the Subnet Manager Node
Run the following as root:
# $HPCX_SHARP_DIR/sbin/sharp_daemons_setup.sh -s -d sharp_am
Daemon's log location is: /var/log/sharp_am.log
Set the "run level".
Start sharp_am as root.
# service sharp_am start
Removing SHARP_am
To remove sharp_am, run the following on the AM host:
# $HPCX_SHARP_DIR/sbin/sharp_daemons_setup.sh -r -d sharp_am
Upgrading NVIDIA SHARP_am
Upgrading SHARP_am requires removal and re-registration as instructed in the sections above.