Re: multi-homing for provider-assigned IPv6 addresses

[[e-mail]]

Chris,

> It seems to me that most use cases for ipv6 multi-homing with provider-assigned addresses only need to route based on source address when the destination prefix is the default route.  So why not require that source prefixes can only be paired up with the default destination prefix ::/0?  That is, we could require that a router can only route based on the source address after the longest match route lookup based on destination address has reached ::/0.  That way, a router can still route packets local to the site based on destination address.  Only once it is clear that the packet is going to the Internet can the router use the source address to pick the correct CE router to get there.
> 
> I see that there was some discussion about this in 2013 on a thread in homenet and rtgwg in the context of draft-baker-rtgwg-src-dst-routing-use-cases, and a number of different opinions were expressed mainly from the point of view of what is needed for homenet.  At this point, RTGWG also needs to consider the impact of proposed solutions on routing scalability.  On the surface, this simplification would seem to improve the scaling properties of the solution in draft-ietf-rtgwg-dst-src-routing, while covering most use cases.

That will not work I'm afraid.
ISPs do not only advertise default routes to CPEs (see RFC6204) but also more specifics (RFC4191).
The more specifics are typically used for walled garden etc.
These will result in (S,D) routes in the network.

> Regardless of how this discussion turns out, I think the draft itself will benefit from more active discussion of the tradeoffs involved.

agreed.

we did a src/dst implementation in VPP (open source software data plane).
the results should be up in the proceedings soon, but until then the slides are here:
https://docs.google.com/presentation/d/1pXJO5vphWFmHAQsrKAgzXYxd-EvMFuxiNYDPXmiMCMk/edit#slide=id.p

in short, back-tracking is quite painful, and if one implemented this as trees of trees, worst case is very bad (128*128). we implemented is a bi-hash with 256 bit key.

a better approach might be to do S lookup first, find the D table, and do the D lookup. that requires duplication of all D,* routes into every D table, but should result in the same behaviour as the D first lookup. worst case then would be 128+128.

there are many alternatives that has been proposed over the years.
e.g.: draft-despres-sam, LISP, ILNP, RFC2260

that is build an overlay from every host/first-hop router to the set of edges.

the problem in the homenet, and I expect in small businesses as well, is that the edge routers are not aware of each other and provided by different entities (the ISPs).

when _you_ say scalability, what do you mean? Multi-prefix multi-homing is designed to alleviate the problem of global route scaling. Within the networks that will do D,S routing, I expect routing tables to be very small. E.g. if you were a company of cisco's size, with tens of providers, you'd do PI multi-homing.

Cheers,
Ole




> 
> Thanks,
> Chris
> 
> -----Original Message-----
> From: Chris Bowers
> Sent: Tuesday, March 15, 2016 8:14 AM
> To: '[e-mail]' <[e-mail]>
> Subject: multi-homing for provider-assigned IPv6 addresses
> 
> RTGWG,
> 
> We scheduled a significant amount of time in Buenos Aires for a discussion of multi-homing for provider-assigned IPv6 addresses for enterprise networks as well as homenet.  I want to explain the motivation for this and provide some background on the topic.  And hopefully spark some discussion on the list before Buenos Aires, as well.
> 
> RTGWG adopted draft-ietf-rtgwg-dst-src-routing last October in the context of supporting multi-homing for provider-assigned IPv6 addresses in homenet.  In Yokohama, the v6ops WG had a lengthy discussion about the need for a solution to support multi-homing for provider-assigned IPv6 addresses for enterprise networks in general (as opposed to just homenet).  This discussion took place in the context of draft-ietf-v6ops-design-choices.  That discussion can be found at:
> 
> https://youtu.be/VzH7yqqGiGc?t=5835
> 
> This discussion led to the email (copied below) from Fred Baker in his role as v6ops co-chair to the chairs of several working groups in which drafts related to this topic are being discussed.
> 
> Our meeting in Buenos Aires includes a 20 minute time slot to discuss the background and motivation of this request from v6ops.  The topic of multi-homing for provider-assigned IPv6 addresses has a long history with many documents written, so I thought it would be useful to highlight a few of the more recent documents that I found to be particularly useful reading.
> 
> RFC 7157 "IPv6 Multihoming without Network Address Translation"
> https://datatracker.ietf.org/doc/rfc7157/
> 
> draft-ietf-6man-multi-homed-host-06 "Routing packets from hosts in a multi-prefix network"
> https://datatracker.ietf.org/doc/draft-ietf-6man-multi-homed-host/
> 
> In addition, it would obviously be useful read or re-read the RTGWG document on this topic:
> https://datatracker.ietf.org/doc/draft-ietf-rtgwg-dst-src-routing/
> 
> There following expired document is also helpful.
> https://www.ietf.org/archive/id/draft-baker-rtgwg-src-dst-routing-use-cases-01.txt
> 
> And finally, this document is useful to understand a concrete proposal for how src/dst routing information could be carried in a link-state routing protocol.
> https://datatracker.ietf.org/doc/draft-baker-ipv6-isis-dst-src-routing/
> 
> I look forward to a fruitful discussion on this topic on the list and in Buenos Aires.
> 
> Thanks,
> Chris
> 
> -----Original Message-----
> From: Fred Baker (fred) [mailto:[e-mail]]
> Sent: Sunday, November 01, 2015 11:03 PM
> To: [e-mail]; [e-mail]; [e-mail]; [e-mail]; 6man Chairs <[e-mail]>; [e-mail]
> Cc: [e-mail] WG <[e-mail]>; The IESG <[e-mail]>
> Subject: PA Address Multihoming in IPv6
> 
> This email is being sent in accordance with the v6ops charter, which calls for the working group to communicate operational issues and requirements to working groups that are chartered to address them.
> 
> The IETF's current primary recommendation for multihoming of midrange enterprise networks - those that cannot justify the costs and overheads of a PI address and in fact multihome - is to obtain a provider-allocated prefix from each of their upstream networks, and deploy a /64 out of each on each LAN in their networks.
> 
> https://tools.ietf.org/html/rfc4213
> 4213 Basic Transition Mechanisms for IPv6 Hosts and Routers. E. Nordmark,
>     R. Gilligan. October 2005. (Format: TXT=58575 bytes) (Obsoletes
>     RFC2893) (Status: PROPOSED STANDARD) (DOI: 10.17487/RFC4213)
> 
> This has a number of issues, not the least of which is in the back end OSS software, which needs to now scale to a much larger number of prefixes, handle multiple addresses in DNS for servers and perhaps clients, resolve reverse DNS queries, and so on. It also is obviously carrying that much more information in routing.
> 
> One outcome of v6ops' discussions this morning was that PI multihoming demonstrably works, but PA multihoming when the upstreams implement BCP 38 filtering requires the deployment of some form of egress routing - source/destination routing in which the traffic using a stated PA source prefix and directed to a remote destination is routed to the provider that allocated the prefix. The IETF currently has no such recommendation, or consensus that it should have. However, enterprise networks are known to delaying operational deployment of IPv6 in part due to the complexities visited upon them and the cost of the back end software upgrades, and this is part of that issue.
> 
> Without trying to limit the options available to the working groups in question, I'll point out that options currently on the table include the following. There are also current open source implementations of source/destination and source-specific routing in IS-IS, OSPFv3, and Babel.
> 
> https://datatracker.ietf.org/doc/draft-baker-ipv6-isis-dst-src-routing
>  "IPv6 Source/Destination Routing using IS-IS", Fred Baker, David
>  Lamparter, 2015-10-19
> 
> https://datatracker.ietf.org/doc/draft-boutier-babel-source-specific
>  "Source-Specific Routing in Babel", Matthieu Boutier, Juliusz
>  Chroboczek, 2015-05-27
> 
> https://datatracker.ietf.org/doc/draft-ietf-6man-multi-homed-host
>  "Routing packets from hosts in a multi-prefix network", Fred Baker,
>  Brian Carpenter, 2015-10-15
> 
> https://datatracker.ietf.org/doc/draft-ietf-ospf-ospfv3-lsa-extend
>  "OSPFv3 LSA Extendibility", Acee Lindem, Sina Mirtorabi, Abhay Roy, Fred
>  Baker, 2015-10-08
> 
> https://datatracker.ietf.org/doc/draft-ietf-rtgwg-dst-src-routing
>  "Destination/Source Routing", David Lamparter, 2015-10-17,
> 
> https://datatracker.ietf.org/doc/draft-sarikaya-6man-sadr-overview
>  "Source Address Dependent Routing and Source Address Selection for IPv6
>  Hosts: Problem Space Overview", Behcet Sarikaya, Mohamed Boucadair,
>  2015-08-17
> 
> https://datatracker.ietf.org/doc/draft-sarikaya-6man-sadr-ra
>  "IPv6 RA Option for Source Address Dependent Routing", Behcet Sarikaya,
>  2015-06-08
> 
> https://datatracker.ietf.org/doc/draft-sarikaya-dhc-6man-dhcpv6-sadr
>  "DHCPv6 Solution for Source Address Dependent Routing", Behcet Sarikaya,
>  2015-05-08
> 
> https://datatracker.ietf.org/doc/draft-xu-ospf-multi-homing-ipv6
>  "Extending OSPFv3 to Support Multi-homing", Mingwei Xu, Shu Yang,
>  Jianping Wu, Fred Baker, 2015-10-11,
> 
> https://datatracker.ietf.org/doc/draft-baker-rtgwg-src-dst-routing-use-cases
> 
> _______________________________________________
> rtgwg mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/rtgwg