Installing UFM Server Software

NVIDIA UFM Enterprise User Manual v6.17.1 Download PDF

The default UFM® installation directory is /opt/ufm.

For instructions on installing the UFM server software, please refer to following instructions per desired installation mode.

The following processes might be interrupted during the installation process:

  • httpd (Apache2 in Ubuntu)

  • dhcpd

Note

To install UFM over static IPv4 configuration (instead of DHCP) please refer to Configuring UFM Over Static IPv4 Address before installation.

After installation:

  1. Activate the software license

  2. Perform initial configuration

Note

Before you run UFM, ensure that all ports used by the UFM server for internal and external communication are open and available. For the list of ports, see Appendix – Used Ports.

Verify that a supported version of Linux is installed on your machine. For details, see UFM System Requirements.

The following table lists the packages that must be installed on your machine (according to the system OS) before you install the UFM server software.

RedHat 7

RedHat 8

RedHat 9

Ubuntu 18.04

Ubuntu 20.04

Ubuntu 22.04

acl

acl

acl

acl

acl

acl

apr-util-openssl

apr-util-openssl

apr-util-openssl

apache2

apache2

apache2

bc

bc

bc

bc

bc

bc

cairo

gnutls

gnutls

chrpath

chrpath

chrpath

gnutls

httpd

httpd

cron

cron

cron

httpd

iptables

iptables-nft

gawk

gawk

gawk

iptables

jansson

jansson

lftp

lftp

lftp

lftp

lftp

lftp

libcurl4

libcurl4

libcurl4

libxml2

libnsl

libnsl

logrotate

logrotate

logrotate

libxslt

libxml2

libxml2

python3

python3

python3

mod_session

libxslt

libxslt

qperf

qperf

qperf

mod_ssl

mod_session

mod_session

rsync

rsync

rsync

net-snmp

mod_ssl

mod_ssl

snmpd

snmpd

snmpd

net-snmp-libs

net-snmp

net-snmp

sqlite3

sqlite3

sqlite3

net-snmp-utils

net-snmp-libs

net-snmp-libs

sshpass

sshpass

sshpass

net-tools

net-snmp-utils

net-snmp-utils

ssl-cert

ssl-cert

ssl-cert

php

net-tools

net-tools

sudo

sudo

sudo

psmisc

php

php

telnet

telnet

telnet

python3

psmisc

psmisc

zip

zip

zip

python3-libs

python36

python3

qperf

qperf

qperf

rsync

rsync

rsync

sqlite

sqlite

sqlite

sshpass

sshpass

sshpass

sudo

sudo

sudo

telnet

telnet

telnet

zip

zip

zip

In addition, ensure the following before you begin installation:

  • The computer hostname is not defined as 127.0.0.1 and localhost is defined as 127.0.0.1.

  • The hostname must NOT appear on the loopback address line. An example of the loopback address is: 127.0.0.1 localhost.localdomain localhost.

  • Disable the firewall service (/etc/init.d/iptables stop), or ensure that the required ports are open (see the prerequisite script, refer to Used Ports).

  • SELinux is disabled.

  • If more than one fabric is managed by different UFM instances, set up different management network spaces for each fabric (not the same LAN).

  • Uninstall any previously installed Subnet Manager from the UFM server machine.

  • MLNX_OFED 5.x version is installed prior to installing UFM.

  • As of UFM v.6.12.0, it is NOT mandatory to configure the IPoIB fabric interface with an IP address.

    In cases where the IP is configured, it is mandatory that the IP is permanently configured and that it starts automatically upon server reboot (the IPoIB fabric interface should be active even if the network is down).

    Note

    The user can set a persistent IP address using Netplan (mainly for Ubuntu systems) or modifying the interface network script (RedHat systems).

  • The default MLNX_OFED installation includes opensm. Remove the MLNX_OFED opensm before UFM installation like the following examples:

    RedHat:

    Copy
    Copied!
                

    rpm -e opensm-3.3.9.MLNX_20111006_e52d5fc-0.1

    Ubuntu:

    Copy
    Copied!
                

    apt purge opensm

    By default, ib0 and eth0 are configured as primary access points for the UFM management. If different management and/or InfiniBand interfaces (including bond interfaces) are used as the primary access points, you should modify the configuration file by running the script /opt/ufm/scripts/change_fabric_config.sh as described in the section Configuring General Settings in gv.cfg.

    Change the UFM Agent interface to the Ethernet and/or IPoIB interfaces used for communication with UFM Agent:

    Copy
    Copied!
                

    ufma_interfaces = ib0,eth0

  • Reliable and high-capacity out-of-band IP connectivity between the UFM Primary and Secondary servers (1 Gb Ethernet is recommended). This connectivity is used for DRBD synchronization.

  • Format two identical servers with dedicated disk partitions for UFM replication. Since the UFM configuration file is replicated to the standby server, both master and standby servers must have the same interfaces.

  • Allocate exactly the same size partition on both servers (master and slave) for the replicated data. See UFM Server Requirements for the recommended partition size.
    Partitions should not be mounted and must be zeroed (the file system should not be installed on the partitions). For disk partitioning, see the Linux user manual (man fdisk).

  • We recommend establishing a passwordless SSH (via /root/.ssh/authorized_keys file) between the two servers before the installation.

  • In fabrics consisting of multiple tiers of switches, it is recommended that the management ports (ib0) of the primary and secondary UFM server be connected to different fabric switches on the same tier (the outermost edge in CLOS 5 designs).

    This is because by default, UFM manages the IB fabric via ib0, port 1 of the HCA. Failure or disconnect of ib0, the IB management port, causes a failure condition in UFM resulting in HA failover.
    When the management ports (ib0) of the primary and secondary UFM server are connected to the same switch, a failure of this switch will result in a disconnect of both UFMs from the fabric, and therefore UFM will not be able to manage the fabric.

Note

Subnet Manager is running over the native InfiniBand layer, therefore bonding the IpoIB interfaces will not provide high availability. For additional information, please refer to section UFM Failover to Another Port.

The UFM installation includes the InfiniBand Performance Management module (IBPM). This module is responsible for reporting performance information back to UFM and upper layer applications. When available, this process is offloaded to the non-management port (default ib1) of the UFM server. Failure or disconnect of the non-management port (ib1) on the primary UFM server will not cause UFM to failover. By default, the UFM Health Monitoring process is configured to try to restart the IBPM. For more information, see UFM Health Configuration in the UFM User Manual.

© Copyright 2024, NVIDIA. Last updated on Jun 18, 2024.