UFM Enterprise 6.16 Bare Metal Installation

Abstract: NVIDIA UFM Enterprise is the centralized management platform for InfiniBand networks, providing topology visualization, performance monitoring, and congestion control. This guide details the deployment process of UFM 6.16 on RHEL 7.9.

1. Prerequisites

1.1 System Requirements

• OS: Red Hat Enterprise Linux Server 7.9 (Maipo).
• Kernel: Default RHEL 7.9 kernel recommended.
• IB Driver: MLNX_OFED_LINUX-23.10-2.1.3.1-rhel7.9-x86_64 (Must match OS exactly).
• UFM Version: ufm-6.16.0.el7.x86_64.mofed5.

1.2 Hardware & Partitioning

• Hardware: Dedicated physical server recommended for UFM Master.
• Partitions:
  • /boot/efi: 1G
  • /boot: 1G
  • swap: 32G (Required by UFM processes)
  • /: Remaining space (Logs are stored in /opt/ufm, ensure sufficient space).

1.3 Environment Setup

1. Network: Static IP configured; note MAC address for licensing.
2. SSH: Passwordless SSH setup between nodes.
3. Yum Repo: Local or network repositories configured.

1.4 Dependencies & Conflict Removal

UFM requires specific Python/Web packages and conflicts with standard OpenSM.

# 1. Install Dependencies
yum install -y sshpass python3 python3-libs python3-mod_wsgi \
    lftp zip rsync telnet qperf net-tools apr-util-openssl \
    mod_ssl mod_session libxslt gnutls net-snmp net-snmp-libs \
    net-snmp-utils httpd psmisc cairo

# 2. Remove Conflicts (OpenSM & Sharp)
# UFM bundles its own optimized OpenSM
yum remove -y sharp-3.1.1*
rpm -e opensm-devel opensm opensm-static opensm-libs

2. Software & License

2.1 Obtain License

Generate the license file using the Master Node's MAC address at the NVIDIA Licensing Portal.

• Path: /opt/mellanox/licenses/ (Create directory if missing).

2.2 Download Package

Upload the installer to /opt/UFMpackage/.

• File: ufm-6.16.0.el7.x86_64.mofed5.tgz

3. Installation Process

3.1 Extract & Run

cd /opt/UFMpackage
tar -zxvf ufm-6.16.0.el7.x86_64.mofed5.tgz
cd ufm-6.16.0.el7.x86_64.mofed5

# Run Installer
./install.sh

3.2 Interactive Setup

Confirm the prompt:

Do you want to install UFM Enterprise [y|n]? y

The script performs automatic checks:

• ✅ Architecture & Distro
• ✅ OFED Driver Compatibility
• ✅ IB Interface Status
• ✅ SELinux Status
• ✅ Disk Space & Permissions

3.3 Completion

Success log:

[*] UFM Installation finished successfully.
[*] To enable UFM on startup run:
    systemctl enable ufm-enterprise.service

HA Deployment Note

If deploying a High Availability (HA) cluster, DO NOT start the service manually at this stage. The service will be managed by the cluster resource manager after HA configuration.

4. Access & Management

4.1 Start Service (Standalone Mode)

For single-node deployments:

# Enable on boot
systemctl enable ufm-enterprise.service

# Start service
systemctl start ufm-enterprise.service

# Check status
systemctl status ufm-enterprise.service

4.2 Web UI Access

• URL: https://<UFM_IP>/ufm
• Default Credentials:
  • User: admin
  • Password: 123456

4.3 Maintenance

• Logs: /opt/ufm/files/logs/ufm-enterprise.log
• Config: /opt/ufm/files/conf/gv.cfg (To change ports).