[rbridge] An ARP/ND draft

[Arien Vijn]

On Jul 18, 2010, at 4:13 AM, Donald Eastlake wrote:

[..]

> In any case, the ability to use "old" cached IP-->MAC mapping as a
> hint for an RBridge to try first when a query comes along seems like
> it could significantly reduce broadcasts.


For what I have seen in practice, the far majority of broadcasted ARP queries are for unreachable hosts cq. unused IP addresses. The broadcast levels are significantly lower for reachable hosts. 

It is the same story with regards to IPv6. The total level of ND messages for unknown addresses is considerably higher than for reachable addresses. However, hosts only have to listen to their own multicast group. All other queries can be ignored by the NICs.

For example, I just looked at one of my routers. It transmitted 1047 IPv6 ND queries over a five minute period. Of which, 1041 queries were for 15 non-existing addresses. So 6 queries would have been optimized and 1041 would have been flooded. 

Hence I think that mapping known IPs to MACs would reduce broadcasts/multicasts due to ARP/ND queries by just a tiny bit. Blocking futile queries would significantly reduce broadcasts/multicasts.

My 2 euro cents.

-- Arien