Title: Unattended Mellanox Infiniband OFED Installation - Tutorial
OS: CentOS Linux
release 7.8.2003 (Core)
Kernel Version:
3.10.0-1127.13.1.el7.x86_64
Compiler: gcc
v4.8.5
Software:
Mellanox OFED
Software Version: 5.0-2.1.8.0
User Account: Use
‘sudo’ (recommended)
Reboot Required:
Yes (recommended)
·
Create a directory ‘/tools/apps/sources/mellanox/ofed/5.0-2.1.8.0’
(Please create directory according to your requirements)
o
mkdir -p
/tools/apps/sources/mellanox/ofed/5.0-2.1.8.0
o export MLNX_ROOT_DIR=/tools/apps/sources/mellanox/ofed/5.0-2.1.8.0
·
Download the latest Mellanox OFED .tgz file for
your OS distribution and architecture from Mellanox website (https://www.mellanox.com/products/infiniband-drivers/linux/mlnx_ofed)
and place it into the directory created in the previous step. In my case the
downloaded file is ‘MLNX_OFED_LINUX-5.0-2.1.8.0-rhel7.8-x86_64.tgz’.
·
Navigate into the directory and extract the tarball.
o
cd ${MLNX_ROOT_DIR}
o
tar xvfz MLNX_OFED_LINUX-5.0-2.1.8.0-rhel7.8-x86_64.tgz
·
Navigate into ‘MLNX_OFED_LINUX-5.0-2.1.8.0-rhel7.8-x86_64’ directory
o
export
MLNX_SOURCE=MLNX_OFED_LINUX-5.0-2.1.8.0-rhel7.8-x86_64
o
cd ${MLNX_ROOT_DIR}/${MLNX_SOURCE}
(If your kernel version is not listed in ‘.supported_kernels’ file then you will have to add support for your
kernel. To accomplish this, please follow the steps under ‘Troubleshooting’
section below)
· Mellanox package comes with some sample configuration files which can be used to install mellanox OFED unattended. Configuration samples can be found in ‘${MLNX_ROOT_DIR}/${MLNX_SOURCE}/docs/conf’ directory i.e ofed-basic.conf, ofed-hpc.conf and ofed-all.conf.
·
Install the necessary packages required for
Mellanox OFED installations
o
yum install
tcl tk
·
To install Mellanox OFED stack run the following
script with the necessary flags for unattended installation. You can use
configuration file according to your requirements.
o ./mlnxofedinstall –c docs/conf/ofed-hpc.conf --force
·
After the above command, the installation will
be completed. Restart the system so that it may load the device drivers on
startup.
How to start infiniband related services?
Once the installation
has been performed, it installs two main services:
·
Openibd: Infiniband related drivers
·
Opensmd: Subnet manager
An instance of a ‘Subnet Manager’ is required on an infiniband fabric (network) for machines to communicate. This service can either be started on Infiniband switches (subject to having the ‘Subnet Manager’ support on the switches) or on servers connected to the same infiniband fabric.
‘openibd’ service needs to be running on all servers whereas ‘opensmd’ service can only be started on a single server/switch or on multiple core servers/switches.
·
Start
and enable ‘openibd’ service
o systemctl start openibd
o systemctl enable openibd
·
Start
and enable ‘opensmd’ service
o systemctl start opensmd
o systemctl enable opensmd
Adding Additional
Kernel Support
Sometimes a downloaded .tgz file doesn’t come
with the RPMs which support the latest kernel. To add support for a different
kernel follow these steps:
o
Navigate into the Mellanox OFED source
directory and run these commands
- cd MLNX_OFED_LINUX-5.0-2.1.8.0-rhel7.8-x86_64
- yum install python-devel
- ./mlnx_add_kernel_support.sh -m . -make-tgz
The above command will re-compile
Mellanox OFED stack and will create a ‘tgz’ file in ‘/tmp’ directory. This will be the tarball which can be used to
install the OFED stack on machines with a specific kernel.
No comments:
Post a Comment