MR access points are supposed to use proxy-arp to respond to ARP requests on behalf of connected clients. However, I've noticed that my MR42s are NOT proxy-arp responding to ARP requests from my MX67. Specifically, an ARP requests from the MX67 are forwarded to the end client, needlessly wasting airtime. Given that the MX67 ARPs for all connected clients every 15 seconds this is generating a lot of ARP requests over the air. Is anyone else seeing the same thing? There's basically two scenarios:
MX sends client-status ARP request for a host connected behind MR42. The MR42 receives this ARP request and forwards it to the end client in question. The client responds and the MR forwards this response to the MX. This is wasting airtime.
Another host on the network sends an ARP request for the same host connected behind MR42. The MR42 receives this ARP request and immediately responds (proxy-arp) without forwarding it to the end client, as expected. This saves airtime.
My MX is on 14.50. The MR42 is on 26.7. This happens with numerous clients and doesn't appear to be client-specific. Running Wireshark on a 2018 Macbook Pro 15'' shows the ARP requests from the MX, but no ARP requests from any other devices. Interestingly, this only happens on ARP requests from the MX. If I swap the MX for my Cisco Firepower1010, configured with the same LAN IPs, etc, the MR correctly sends proxy-arp responses.
Replying to my own thread here. I think I figured out the difference. The ARP requests from the MX are unicast ARP requests, not standard broadcast ARP requests. Proxy-ARP on the MR42 seems to only handle broadcast ARP requests, not client-status unicast ARP requests. In this case, it technically isn't wasting airtime because it isn't having to broadcast ARP to all connected clients. I'd still prefer if the MR replied to it on behalf of the clients given that the MX sends an ARP request every 15 seconds.