Large Scale Q-in-Q

1
Large Scale Q-in-Q
1. Scaling address learning mick_seaman_at_ieee.or
g
2
Perceived problem

• As a Provider Bridged Network grows and attracts
  more customers it will contain more Provider
  Bridges that learn the addresses of each customer
  so the total amount of address learning
• fn(of customers x of Provider Bridges)
• fn(of customers2)
• fn(value provided2)

3
Preliminary answer

• Dont learn on point-to-point Service Instances
• But what about Services Instances with more than
  2 points of attachment?

4
Better answer

• Dont learn unless the learning will affect where
  frames are forwarded e.g. dont learn when
• Receiving port is attached to point-to-point LAN
• There are less than two ports on this bridge that
  are in the (egress) Member set for this VLAN

5
Making the best of this

• Use GVRP to scope the extent of the Service VLAN
  so extent of VLAN is minimal over its Spanning
  Tree but extent of VLAN autoconfigures after
  active topology reconfiguration
• For S-VLAN with 3 customer attachments learning
  only required in 1 provider bridge
• For S-VLAN with n customer attachments learning
  only required in (n-2) provider bridges at most
• Minimizes database flushing impact after change

6
Shared VLANs

• Some stations transmit on A, receive on B
• Others receive on A, transmit on B
• Allows multiple subscribers to be served using
  one customer network

7
Optimizing learning on shared VLANs

• Only learn addresses from frames on VLAN A if
  VLAN B has more than one way for a frame that has
  entered a bridge to leave

8
LARGE scale with shared VLANs
Routers/servers here
Addresses of subscribers (below) only learnt in
this bridge and below
Subscribers attached to these bridges