Hi co-authors of the draft,

I have read your draft 
(http://tools.ietf.org/id/draft-bonica-l3vpn-orf-covering-prefixes-00.txt) and 
feel the CP-ORF described in the draft may be applicable in Virtual Subnet 
(http://tools.ietf.org/html/draft-xu-l3vpn-virtual-subnet-01), a special 
BGP/MPLS IP VPN [RFC4364] application environment as well, to realize the 
on-demand RIB installation (a.k.a., on-demand host route pull) purpose (see 
section 3.6.3. PE Router RIB Reduction at 
http://tools.ietf.org/html/draft-xu-l3vpn-virtual-subnet-01#page-12).

However, to use the CP-ORF in the Virtual Subnet environment, there is a 
practical issue which needs to be considered. To illustrate the issue clearly, 
I would use the Figure 6: RIB Reduction Example described in the Virtual Subnet 
draft as an example. Since the V-spokes (i.e., PE-1 and PE-2 in Figure 6) have 
already installed a directly connected route (i.e., 1.1.1.0/24), the VPN IP 
default route which is expected to be originated by the V-hub (i.e., the RR in 
Figure 6) would not be applicable anymore. For example, when PE-1 receives a 
packet destined for CE host B (i.e., 1.1.1.3) from its local CE host (e.g., 
host A), PE-1 would not be able to forward that packet to the RR (i.e., the 
V-hub) since the longest match route for that packet is 1.1.1.0/24, rather than 
the default route. Therefore, it would be desirable to allow the V-hub to 
advertise two more specific routes (e.g., 1.1.1.0/25 and 1.1.1.128/25) rather 
than a default route in this case. Of course, this behavior may conflict with 
the original concept of the Virtual Hub-and-Spoke VPN 
[I-D.ietf-l3vpn-virtual-hub] where the V-hub is required to originate a VPN IP 
default route. I wonder whether it would be better to relax that constrain, 
e.g., the V-hub is only required to originate a default route or one or more 
generic routes which cover the more specific routes to be suppressed. In this 
way, the case where V-spokes has default routes (see Figure 2 to 4 in the 
Virtual Subnet draft) can be easily addressed without using the complex tricks 
described in Section 5 of Virtual Hub-and-Spoke VPN draft (see 
http://tools.ietf.org/html/draft-ietf-l3vpn-virtual-hub-08#page-10). In other 
word, the default route could be originated by the V-spokes as normal and that 
route would be installed by all the other V-spokes and even the V-hub. In this 
way, the inter-subnet traffic which uses the default route would be forwarded 
along the optimal path.

To achieve the above goal, the following rule stated in the CP-ORF draft as " 
the route is more specific than a /64 (i.e., the route more specific than an IP 
VPN default route) should be replace with "the prefix length of the route is 
larger than or equal to the value of "Min-Prefix-Length" field in the CP-ORF". 
In other word, a "Min-Prefix-Length" field should be added to the existing 
CP-ORF Type-specific Encoding.

Best regards,
Xiaohu  

Reply via email to