Re: About the WG adoption of draft-xu-l3vpn-virtual-subnet-fib-reduction-00

["Henderickx, Wim (Wim)" <[e-mail]>]

See right now there is /32 or /128 routes which can come from hosts in
virtual subnet/CE-PE protocol/static routes/loopbacks/etc and to figure
out the origin you have very little means doing so. So if an operator
wants to distinguish the installation by default in the FIB it would be
good to tag the origin and he can than decide based on local policy what
to do. I believe it is a good practice and should be described in the
draft.

This is what I meant to say

On 30/07/14 11:48, "Xuxiaohu" <[e-mail]> wrote:

>Hi Wim,
>
>I just want to figure out in which scenario the host route to the VRF
>interface address of a remote PE router should be FIB-installed by
>default. If there is a scenario, I agree that your proposal of using
>communities is the best way.
>
>Best regards,
>Xiaohu
>
>> -----Original Message-----
>> From: Henderickx, Wim (Wim) [mailto:[e-mail]]
>> Sent: Wednesday, July 30, 2014 5:37 PM
>> To: Xuxiaohu; [e-mail]
>> Subject: Re: About the WG adoption of
>> draft-xu-l3vpn-virtual-subnet-fib-reduction-00
>> 
>> In that case you can argue why not do it for all routes as well.
>> 
>> On 30/07/14 11:30, "Xuxiaohu" <[e-mail]> wrote:
>> 
>> >Why do you need to distinguish them from each other? In other words,
>> >why can't PE routers process the host route to a given VRF loopback
>> >address of a remote PE router as a normal remote host route (i.e.,
>> >on-demand installation)?
>> >
>> >Best regards,
>> >Xiaohu
>> >
>> >-----Original Message-----
>> >From: Henderickx, Wim (Wim) [mailto:[e-mail]]
>> >Sent: Wednesday, July 30, 2014 5:20 PM
>> >To: Xuxiaohu; [e-mail]
>> >Subject: Re: About the WG adoption of
>> >draft-xu-l3vpn-virtual-subnet-fib-reduction-00
>> >
>> >Why not, if I configure a loopback in the VRF it has to be advertised,
>> >this is generic VRF functionality, nothing to do with virtual subnet. I
>> >am not talking to the IP address on the virtual subnet interface.
>> >In this case you need to distinguish between host routes and these
>>VRFs.
>> >Using communities is the best way afais.
>> >
>> >On 30/07/14 10:59, "Xuxiaohu" <[e-mail]> wrote:
>> >
>> >>Hi Wim,
>> >>
>> >>In the Virtual Subnet context, the host route corresponding to the VRF
>> >>interface address doesn't need to be advertised.
>> >>
>> >>Best regards,
>> >>Xiaohu
>> >>
>> >>> -----Original Message-----
>> >>> From: Henderickx, Wim (Wim)
>> >>> [mailto:[e-mail]]
>> >>> Sent: Wednesday, July 30, 2014 4:55 PM
>> >>> To: Xuxiaohu; [e-mail]
>> >>> Subject: Re: About the WG adoption of
>> >>> draft-xu-l3vpn-virtual-subnet-fib-reduction-00
>> >>>
>> >>> Real loopbacks. E.g. A loopback /32 or /128 configured in the VRF
>> >>>
>> >>> On 30/07/14 10:52, "Xuxiaohu" <[e-mail]> wrote:
>> >>>
>> >>> >Hi Wim,
>> >>> >
>> >>> >Did you mean PE's loopback addresses by "real loopbacks"?
>> >>> >
>> >>> >Best regards,
>> >>> >Xiaohu
>> >>> >
>> >>> >> -----Original Message-----
>> >>> >> From: Henderickx, Wim (Wim)
>> >>> >> [mailto:[e-mail]]
>> >>> >> Sent: Wednesday, July 30, 2014 4:31 PM
>> >>> >> To: Xuxiaohu; [e-mail]
>> >>> >> Subject: Re: About the WG adoption of
>> >>> >> draft-xu-l3vpn-virtual-subnet-fib-reduction-00
>> >>> >>
>> >>> >> What I would like to see is a way to identify the host routes
>> >>> >>since there are 2
>> >>> >> levels: real loopbacks that need to be installed by default and
>> >>> >>real host routes  that can be installed on demand. It would be
>> >>> >>good to show how the control  plane could distinguish them using
>> >>> >>communities or the likes.
>> >>> >>
>> >>> >> On 30/07/14 08:54, "Xuxiaohu" <[e-mail]> wrote:
>> >>> >>
>> >>> >> >Hi all,
>> >>> >> >
>> >>> >> >Virtual Subnet
>> >>> >> >(http://tools.ietf.org/html/draft-ietf-l3vpn-virtual-subnet) is
>> >>> >> >intended for building L3 network virtualization overlays within
>> >>> >> >and/or across data centers. Since a subnet is extended across
>> >>> >> >multiple PE routers, CE host routes need to be exchanged among
>> >>> >> >PE routers. As a result, the forwarding table size of PE routers
>> >>> >> >(e.g., some old ToR
>> >>> >> >switches) may become a big concern in large-scale data center
>> >>> >> >environments. In fact, some folks had already expressed their
>> >>> >> >concerns about this potential FIB scaling issue during the WG
>> >>> >> >adoption poll of
>> >>> >>the Virtual
>> >>> >> Subnet draft.
>> >>> >> >
>> >>> >> >As CE host routes may still need to be maintained on the control
>> >>> >> >plane of PE routers in some cases (e.g.. MVPN scenario), this
>> >>> >> >draft
>> >>> >> >(http://tools.ietf.org/html/draft-xu-l3vpn-virtual-subnet-fib-re
>> >>> >> >d
>> >>> >> >uct
>> >>> >> >ion
>> >>> >> >-00
>> >>> >> >) proposes a very simple mechanism for reducing the FIB size of
>> >>> >> >PE routers without any change to the RIB and even the routing
>> >>>table.
>> >>> >> >
>> >>> >> >During the L3VPN WG session at Toronto, many people had
>> >>> >> >expressed their supports for the WG adoption of this work
>> >>> >> >(Thanks a lot for your supports). However, there are still a few
>> >>> >> >people who are not in favor of the WG adoption. According to WG
>> co-chairs'
>> >>> >> >suggestion, I would like to request those opposers to explain
>> >>> >> >their reasons so that we could further improve the draft if
>> >>>possible.
>> >>> >> >
>> >>> >> >Best regards,
>> >>> >> >Xiaohu (on behalf of all co-authors)
>> >>> >> >
>> >>> >
>> >>
>> >
>