EFA is supported on multiple management IP networks, providing access to EFA from
servers that are on different subnets.
Overview
The Multiple Management IP Network feature offers the following support:
Supports single node and multi-node deployments
Supports TPVM deployments and server and VM-based deployments
You can add additional management networks during EFA installation and after
EFA installation.
You can delete management networks after EFA installation.
The configuration of the multiple networks is migrated during all EFA
upgrade scenarios: single node to single node, single node to multi-node,
and multi-node to single node.
If you do not need multiple
management networks, simply reply "no" when prompted during EFA installation
or upgrade. For instructions, see the installation and upgrade topics in the
Extreme Fabric Automation Deployment
Guide, 2.4.0.
The feature supports up to 6 networks.
The RMA, backup, restore, and upgrade functions are all supported with the
multiple management IP feature.
Assumptions
In a multi-node deployment,
the sub-interface with the VLAN is created under the same NIC as the VIP
destination. In a single-node deployment, the sub-interface is created under
the NIC that you specified as the host IP installation (if there are
multiple NICs). Creating sub-interfaces on different NICs of the server is
not supported.
EFA does not validate to the
newer IP subnets. You are responsible for ensuring reachability.
Changing IP subnets is not
supported. To make any modifications to the IP subnet or VLAN or name, you
must delete the IP subnet and add a new one.
Because of the EFA-wide
changes required by this feature (for example, updates to the Ingress
controller), you can expect about 20 to 30 seconds of downtime during add
and delete operations.
In a high-availability
deployment, both nodes have to be up and available during add and delete
operations (because sub-interface creation and keepalived changes are unique
to the node). Because this is an infrequent operation, you should verify
that both the nodes are up and in READY state before beginning this
operation.