1. Virtual Subnet : A L3VPN-based Subnet Extension Solution
draft-xu-l3vpn-virtual-subnet-01
Xiaohu Xu (Huawei)
Susan Hares (Adara Networks)
Yongbing Fan (China Telecom)
Christian Jacquenet (Orange)
Truman Boyes (Bloomberg LP)
Brendan Fee (Extreme Networks)
IETF88, Vancouver

2. Virtual Subnet at a Glance: Control Plane
Prefix Next-hop Protocol / Direct
/ Direct
/ PE BGP
/ Direct
VPN Subnet: /24
Host B:
/24
MPLS/IP Backbone
PE-1
DC #1
PE-2
Prefix Next-hop Protocol
/ Direct
/ PE BGP
/ Direct
/ Direct
Host A:
/24
VRF:
DC #2
Host route distribution
via L3VPN signaling
Host discovery
based on ARP/ND/VDP etc. 1 3 2
Host route
creation for
local hosts.
/24
Local CE hosts are discovered based on ARP/ND, VDP or even the interaction with the data center orchestration system.
Host routes for local CE hosts are created automatically on PE routers and then propagated to remote PE routers via L3VPN signaling.

3. Virtual Subnet at a Glance: Data Plane
VPN Subnet: /24
Host B:
/24
PE-1
DC #1
PE-2
Host A:
/24
IP(A)->IP(B)
VPN Label
Tunnel to PE-2
VLAN ID
MAC(PE-2)->MAC(B)
DC #2 2
Route
look-up 3 4
ARP
Proxy
MAC(A)->MAC(PE-1) 1
Prefix Next-hop Protocol
/ Direct
/ PE BGP
/ Direct
/ Direct
Prefix Next-hop Protocol / Direct
/ Direct
/ PE BGP
/ Direct
VRF:
/24 5
MPLS/IP Backbone
IP MAC
IP(B) MAC(PE-1)
ARP Table:
PE routers acting as ARP proxies respond with their own MAC addresses to the ARP requests messages for remote CE hosts from local CE hosts.
Intra-subnet traffic across data centers is forwarded according to the L3VPN forwarding process.

4. VM (CE Host) Mobility
VPN Subnet: /24
Host B
/24
PE-1
PE-2
Host A
/24
DC #1
DC #2
Gratuitous ARP 1
BGP route update
for host C 3
IP(C)->MAC(PE-1) 5
Create local host route 2
Update host route 4
Host C moves from
DC#1 to DC#2
ARP
Proxy
Host C
/24
MPLS/IP Backbone
/24
/24
The new PE router (i.e., PE-2) advertises a host route for the arriving VM upon receiving a notification of VM attachment (e.g., a gratuitous ARP ).
The old PE router (i.e., PE-1) withdraws the host route for the moved VM after noticing the leave of that VM. Meanwhile, it would broadcast a gratuitous ARP on behalf of that CE host with source MAC address being one of its own.

5. Confine MAC Learning, Flooding and Failure Domains
VPN Subnet: /24
MPLS/IP Backbone
MAC learning,
Flooding and Failure
domain #1
DC #1
DC #2
IP(A)->IP(B)
VLAN ID
MAC(PE-2)->MAC(B)
Host D
Host B
Host C
Host A
MAC(A)->MAC(PE-1)
PE-1
ARP
Proxy
PE-2 DC
Switch
MAC learning,
Flooding and Failure
domain #2
MAC learning domain is confined within data centers. Therefore, switches within data centers only need to learn MAC addresses of local CE hosts.
Flooding and failure domains are confined within data centers. As such, multicast/broadcast protocol messages (e.g., ARP/DHCP/IGMP/STP/VRRP ) from customer networks are terminated on PE routers. In addition, no flood of unknown unicast across data centers.

6. Optimal Forwarding for North-South Traffic
VPN Subnet: /24
DC #1
DC #2
ARP
Proxy
PE-3
PE-1
PE-2
Cloud User
Prefix Next-hop Protocol
/ PE BGP
/ PE BGP
/ Direct
/ Direct
VRF:
Host B:
/24
GW:
Host A:
/24
MPLS/IP Backbone
VPN Subnet: /24
VRF:
Prefix Next-hop Protocol / Direct
/ Direct
/ PE BGP
/ Direct
/ PE BGP
Prefix Next-hop Protocol
/ Direct
/ PE BGP
/ Direct
/ Direct
/ PE BGP
/24
/24
Host routes for CE hosts within data centers are propagated to remote PE routers to which cloud users are connected. Therefore, north-to-south traffic would be delivered to the right data center without traffic tromboning.
PE routers of each data center are default GWs. Therefore, south-to-north traffic would be forwarded to cloud users without traffic tromboning as well.

7. Considerations for Non-IP traffic
Virtual Subnet is a Layer3 overlay in which IP traffic including both intra- subnet and inter-subnet would be forwarded at Layer3.
To support non-IP traffic further, the unified L2/L3 overlay approach following the idea of “route all IP traffic, bridge non-IP traffic” could be considered (e.g., IP traffic is forwarded by using the Virtual Subnet while non-IP traffic is forwarded across Layer2 overlays (e.g., VPLS) .

8. Next Steps Implementations are already available.
We co-authors believe this draft is ready for WG adoption.