Control-Plane-VXLAN
In this mode; there is no need for the IP Multicast in the underlay transport network; for any traffic that would require to be sent to all VTEPs (like Broadcast or Multicast); the head-end replication will be used instead (as the previous modes)
Dealing with the Unknown Unicast traffic is what really differentiates this mode of operation from the previous modes. In this mode, a ‘Control-Plane’ does exist to distribute the MAC-to-VTEP mapping entries between the different VTEPs, hence no need for any data-plane learning technique (like flood & learn).
This control plane piece could be a Controller (like VMware NSX, Midokura, Nuage, Openstack…), a signaling protocol (like MP-BGP in the EVPN-based VXLAN) or the proposed Blockchain-VXLAN (as will be detailed in the next article).
Controller-based-VXLAN
In the Controller-based VXLAN service, Data-Plane (flow-based) learning is optional or even not needed: the controller synchronizes all the MAC addresses as soon as the different switches learn them from their local ports.
For example in Figure-1, Leaf-1 learns the MAC address of Server-1 from its local port Eth1. This information is automatically and immediately synchronized to the controller that in turn pushes that info to Leaf-2, Leaf-3 & any other VTEP in the same VXLAN domain.
This VXLAN operation depends on the distribution of all learnt MAC addresses from the different VTEPs via the controller that ‘pushes’ to all VTEPs a complete (and always updated) list of MAC-to-VTEP mapping entries.
Because of that, for this mode; there would be no Unknown Unicast as the list of all communicating MACs is on each VTEP, but in case of an unknown MAC (maybe for a destination outside the local VXLAN domain) & depending on the configuration; the local VTEP can direct it via the default entry towards the VXLAN gateway.
For the other Broadcast & Multicast traffic; the head-end replication is always the solution.
EVPN-VXLAN
In the EVPN-VXLAN; each VTEP is now a PE (Provider Edge) node & will learn the local MAC addresses associated to its VXLANs from its local ports as usual.
Using MP-BGP EVPN address family; these entries will be propagated between the different PEs (ideally through a set of MP-BGP RR ‘Route Reflectors’)
As in the Controller-based VXLAN; there would be no Unknown Unicast as the list of all communicating MACs is on each VTEP, but in case of an unknown MAC; the local VTEP can direct the traffic via the default entry towards the VXLAN gateway.
Again for the other Broadcast & Multicast traffic; the head-end replication is always the solution.
To be cont’d
