Installation Notes
Supported NVIDIA Externally Managed Switches
|
Type |
Model |
Latest Tested Firmware Version |
|
NDR switches |
|
31.2021.4036 |
|
HDR switches |
|
27.2012.4036 |
|
EDR switches |
|
15.2010.4402 |
Supported NVIDIA Internally Managed Switches
|
Type |
Model |
Latest Tested OS Version |
|
XDR switches |
|
NVOS-25.02.1002 |
|
NDR switches |
|
MLNX-OS 3.12.1002 NVOS 25.01.4000 |
|
HDR switches |
|
MLNX-OS 3.12.1002 |
|
EDR switches |
|
MLNX-OS 3.10.4400 |
Bare Metal Deployment Requirements
|
Platform |
Type and Version |
|
OS (Relevant for Standalone and High-Availability deployments) |
64-bit OS:
|
|
CPU(a) |
x86_64 |
|
HCAs |
|
|
OFED(b) |
|
(a) CPU requirements refer to resources consumed by UFM. You can also dedicate a subset of cores on a multicore server. For example, 4 cores for UFM on a 16-core server.
(b) For supported HCAs in each MLNX_OFED version, please refer to MLNX_OFED Release Notes.
(c)UFM v6.15.0 is the last version to support NVIDIA ConnectX-4 adapter cards
For running SHARP Aggregation Manager within UFM, it is recommended to use MLNX_OFED-5.4.X version or newer.
Installation of UFM on minimal OS distribution is not supported.
UFM does not support systems in which NetworkManager service is enabled.
Before installing UFM on RedHat OS, make sure to disable the service.
Docker Installation Requirements
UFM Docker Container is supported on the standard docker environment (engine).
The following operating systems were tested with Docker Container (as standalone container):
|
Component |
Type and Version |
|
Supported OS |
|
For UFM Docker Container installation in HA mode, please refer to Bare Metal Deployment Requirements for the list of operating systems and kernels which support HA.
On some Ubuntu OSs, Docker is installed via SNAP, which might lead to errors when trying to use UFM Plugins.
To solve this issue, perform the following:
Remove Docker installed via SNAP, run:
snap remove --purge docker
Update the local package index, run :
apt update
Install native Docker, run:
apt install-y docker.io
Podman Installation
Podman is tested on Oracle Linux and RedHat distributions.
Installation Instructions
Since Docker and Podman are mutually exclusive, ensure that any existing Docker packages are removed before installing Podman. You can do this with the following command:
yum remove $(rpm -qa | grep docker)
Once Docker is removed, proceed with the installation of Podman and the Podman Docker-compatible CLI using the following commands:
dnf install podman podman-docker
Finally, enable and start the Podman socket to allow for socket-based communication:
systemctl enable --now podman.socket
After installing Podman, the deployment of UFM Docker Containers and UFM plugins should function in the same way as the corresponding Docker deployments.
UFM Server Resource Requirements per Cluster Size
|
Fabric Size |
CPU Requirements* |
Memory Requirements |
Disk Space Requirements |
|
|
Minimum |
Recommended |
|||
|
Up to 1000 nodes |
4-core server |
4 GB |
20 GB |
50 GB |
|
1000-5000 nodes |
8-core server |
16 GB |
40 GB |
120 GB |
|
5000-10000 nodes |
16-core server |
32 GB |
80 GB |
160 GB |
|
Above 10000 nodes |
Contact NVIDIA Support |
|||
UFM GUI Client Requirements
The platform and GUI requirements are detailed in the following tables:
|
Platform |
Details |
|
Browser |
Edge, Internet Explorer, Firefox, Chrome, Opera, Safari |
|
Memory |
|
MFT Package Version
|
Platform |
Details |
|
MFT |
Integrated with MFT version 4.30.1-8 |
UFM SM Version
|
Platform |
Type and Version |
|
SM |
UFM package includes SM version 5.21.0 |
Assuming the SM is connected to the production cluster, it can handle any events (IB traps) coming from the fabric that is being built; such events should not affect the routing on the production cluster. If events occurred in the production cluster, the routing could be changed.
However, NVIDIA recommends isolating fabric sections to allow faster bring-ups, faster troubleshooting and misconfiguration avoidance that can cause routing errors. Isolation provides clearer SM and CollectX logs, avoiding warnings/errors from masking real production issues.
UFM NVIDIA SHARP Software Version
|
Platform |
Type and Version |
|
NVIDIA® Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)™ |
UFM package includes NVIDIA SHARP software version 3.9.0 |
Used Ports by UFM Server
For a list of ports used by the UFM Server for internal and external communication, refer to Appendix – Used Ports.
The installer detects versions previously installed on the machine and prompts you to run a clean install of the new version or to upgrade while keeping user data and configuration unchanged.
The upgrade from previous versions maintains the existing database and configuration, allowing a seamless upgrade process.
Upgrading UFM Enterprise software version is supported up to two previous GA software versions (GA -1 or -2).
For example, if you wish to upgrade to UFM Enterprise v6.17.0, it is possible to do so only from UFM Enterprise v6.16.0 or v6.15.0.
Due to a possible conflict, SM and SHARP installed by the MLNX_OFED must be uninstalled. The installation procedure will detect and print all MLNX_OFED packages that must be removed.