NVIDIA UFM-SDN Appliance User Manual v4.15.0
NVIDIA UFM-SDN Appliance User Manual v4.15.0

Changing UFM Running Mode

Warning

In Management or High Availability modes, ensure that all Subnet Managers in the fabric are disabled before running UFM. Any remaining active Subnet Managers will prevent UFM from running.

To change the UFM running mode:

Procedure_Heading_Icon-version-1-modificationdate-1707309326297-api-v2.PNG

  1. Log into the CLI as admin and use the selected password.

  2. Change the running mode.

    Copy
    Copied!
                

    ufm-appliance > enable ufm-appliance # configure terminal ufm-appliance [ mgmt-sa ] (config) # ufm mode <mode>

    UFM running mode:

  • mgmt. - management mode

  • mon - monitoring mode

  • mgmt.-allow-sm - management mode with "External SMs" in the fabric

  • remote-mon - connects UFM-SDN Appliance to the Remote Fabric Collector

  • sm-only - make appliance work as an "External SM" machine
    For further information, please refer to the UFM-SDN Appliance Command Reference Guide.

Run UFM in Monitoring mode while running concurrent instances of Subnet Manager on Mellanox switches. Monitoring and event management capabilities are enabled in this mode. UFM non-monitoring features such as provisioning and performance optimization are disabled in this mode.

The following table describes whether features are enabled or disabled in Monitoring mode.

Features Enabled/Disabled in Monitoring Mode

Feature

Enabled/Disabled in Monitoring Mode

Fabric Discovery

Enabled

Topology Map

Enabled

Fabric Dashboard

Enabled

Fabric Monitoring

Enabled

Alerts and Thresholds (inc. SNMP traps)

Enabled

Fabric Logical Model

Enabled

Device Management

Enabled

Subnet Manager and plugins

Disabled

Subnet Manager Configuration

Disabled

Automatic Fabric Partitioning

Disabled

Central Device Management

Disabled

Quality of Service

Disabled

Failover (High Availability mode)

Disabled

Traffic Aware Routing Algorithm

Disabled

Fabric Collective Accelerator (FCA)

Disabled

Integration with Schedulers

Disabled

Unhealthy Ports

Disabled

In Monitoring mode, UFM periodically discovers the fabric and updates the topology maps and database.

For Monitoring mode, connect UFM to the fabric using port ib0 only. The fabric must have a subnet manager (SM) running on it (on another UFM, HBSM, or switch SM).

Warning

When UFM is running in Monitoring mode, the internal OpenSM is not sensitive to changes in OpenSM configuration (opensm.conf).

Warning

When running in Monitoring mode, the following parameters are automatically overwritten in the conf/opensm/opensm_mon.conf file on startup:

  • event_plugin_name osmufmpi

  • event_plugin_options --vendinfo -m 0

Any other configuration is not valid for Monitoring mode.

  1. Stop the UFM service. Run:

    Copy
    Copied!
                

    ufm-appliance [ mgmt-sa ] (config) # no ufm start

  2. Change the UFM mode. Run:

    Copy
    Copied!
                

    ufm-appliance [ mgmt-sa ] (config) # ufm mode mon

  3. (Optional) In the conf/gv.cfg configuration file:

    • If desired, change mon_mode_discovery_period (the default is 60 seconds)

    • Set reset_mode to no_reset

      It is recommend to use this setting when running multiple instances of UFM so that each port counter is not reset by different UFM instances. For more information, see Resetting Physical Port Counters.

  4. Start the UFM service. Run:

    Copy
    Copied!
                

    ufm-appliance [ mgmt-sa ] (config) # ufm start

    The running mode is set to Monitoring, and the frequency of fabric discovery is updated according to the setting of mon_mode_discovery_period.

Note that a monitor icon will appear at the top of the navigation bar indicating that monitoring mode is enabled:

image2019-9-18_14-45-12-version-1-modificationdate-1707309325567-api-v2.png

© Copyright 2023, NVIDIA. Last updated on Feb 8, 2024.