When it comes to multi domain or Inter datacenter communication, minimizing the broadcast traffic between the datacenters is an important scaling requirement.
Especially if you are dealing with millions of end hosts, localizing the broadcast traffic is critical to save resources on the network and the end hosts. Resources are bandwidth , CPU , memory and so on.
In this post I will mention how ARP cache is populated in OTV and EVPN technologies and the importance of ARP proxy function.
Classical approach to control broadcast traffic by localizing it within a datacenter is Proxying.
ARP is a good example of broadcast packet and ARP Proxy or Proxy ARP works either based on control or data plane learning.
Idea is, destination MAC address can be learned from the local device which keeps ARP cache and ARP traffic doesn’t have to traffic over datacenter interconnect links.
I said ARP cache can be populated either via control or data plane learning and let me give an example for each one of them.
OTV as a Cisco preparatory protocol advertise the MAC addresses through IS-IS. MAC reachability information is learned via control plane. But OTV doesn’t advertise MAC to IP binding through IS-IS. MAC to IP binding is used to populate ARP cache and it is done via data plane learning in OTV.
If MAC to IP binding information would be sent in a control plane protocol, then we would say the ARP cache is populated via control plane. It is not the case in OTV.
On the other hand, EVPN which uses BGP control plane to advertise MAC address reachability information also send IP to MAC binding information through BGP updates.
Thus in EVPN, not only the MAC addresses but also the ARP cache is populated via control plane.
Probably this is one of the biggest misunderstanding on ARP proxy function with two popular Interdomain Layer 2 Overlay protocols.