Hi
I want to make a metro ethernet style design with two ASR9001 to do subscriber L3 termination. However I see limitations in the amount of subinterfaces I can create on a bundle-eth.
The outline of the design:
- Subscribers are connected to a low-end switch (i.e. SG500X) and each subscriber-facing interface will be a QinQ access interface. In this way I have a C-VLAN per subscriber.
- All access-switches will be connected to a pair of aggregation switches. The aggregation switches do QinQ, providing an S-VLAN for each access-switch.
- At the ASR9001s i want a subinterface per subscriber (bundle-eth1.<s-vlan><c-vlan>)
- All subinterfaces are part of the same bridge-domain with split-horizon, DHCP snoop, IP source-guard, ARP inspection, IGMP snoop, MLD snoop etc.
- L3 forwarding is done via BVI interface.
- BVI interface does DHCP relay
- HSRP is used for FHRP between BVI interfaces of both ASR9001s.
The setup will only have to scale for up to 10K subscribers, which seems a bit small to contemplate BNG.
I just tried running this setup through a lab setup and find a limitation of 4096 L2 subinterfaces per trunk:
interface Bundle-Ether100.20053999 l2transport
!!% The L2 sub-interface limit for the trunk interface has been reached: Trunk limit for L2 subinterfaces on Bundle-Ether100 is 4096
I cannot use "encapsulation ambiguous" on L2 subinterfaces, so is there some other way to make it more scalable?