We are running (currently in the lab) dual stacked IPoE based BNG on ASR9k RSP880 running 6.4.2
Subscribers are connected via L2 network to the BNG
We are using ambiguous vlans, and BNG box also acts as DHCP proxy for v4 and v6
It overall works fine, the question is about IPv6 scaling.
For v4 we configured 'arp learning disable' and 'subscriber arp scale-mode-enable'
For v6, what are the scaling recommendations ?
Can we disable 'ipv6 nd' on the Bundle & Dynamic template, and neighbour learning for the individual subscribers, as the forwarding info is in the DHCPv6 proxy data, and therefore I dont need it in the neighbour table too (same reasoning as for v4), and im concerned about scaling of the whole solution. Currently I see the subsciber.ip<no> in my 'show ipv6 neighbour' output, which I guess I dont need there.
Any ideas or pointers appeciated
In the dynamic template can you try out the following configurations:
ipv6 nd suppress-cache-learning
You can also clear the neighbors in the ND cache using "clear ipv6 neighbours"
Can you give us a sample o/p of the show ipv6 neighbours to check .
What is the impact of having this entry in the show ipv6 neighbours and how do you think this is affecting functionality for scaling
Kindly note the configuration shared will help to avoid processing incoming NA messages. This will not stop the incoming NA messages. Only processing of the NA messages will be skipped.
For the main interfaces like bundle ether etc though i have not tested it there is a configuration for setting the cache limit. You can go to the config mode and enter the interface level and configure "ipv6 nd cache-limit 0". From the config it looks like the interface will not move an incomplete entry to complete when this is configured