Hi,
I'm currently facing to an environment where a VPLS has been deployed as a "HUB" in the core of a campus LAN Network, not participating in Spanning-Tree
The design principles are :
- a root VSS is connected to a ASR920 VPLS Network, thanks to an Active/Passive MC-LAG
- each switching zone is connected to two ASR920 to ensure redundancy
- all Vlans (1-4094) are always propagated to all ports of all zones (no L2 segmentation between zones), Vlan tag is not decapsulated in the VPLS and all Vlans stays encapsulated in a single bridge-domain on each ASR920
Here is a generic diagram, not describing all the zones nor the entire infrastructure :
As you can guess, we are facing a lot of looping issues in this kind of design where the VPLS is seen as a "Hub", as soon as a topology change occurs.
I'm currently studying the workarounds with minimum impact on the client L2/L3 Design, and I'm currently thinking to the first following solution :
- Enable PVST/RPVST on ASR920 in order to make the VPLS participating in Spanning-Tree and then acting as a switch and no more as a simple Hub, by :
- enable PVST or RPVST on the ASR920, with a lower priority than VSS Root switches
- swtich EVC service instance to trunk (service instance trunk)
- enable stp peering on each connected port to a switch
- using command to instantiate 1 bridge-domain per Vlan (bridge-domain from-encapsulation)
- an alternative could also be to enable MSTP as VPLS has a single topology, but will be constrained to switch to STP-compatible mode on edges as PVST/RPVST are used on other switches
What do you think of this solution ? Have you ever faced to this kind of design ?
Is there any other better alternative solution with minimal impact on L2/L3 Design ?
Thanks in advance,