Re: [v6ops] Question on multi-homed nodes and address/route selection

[Simon Hobson <[e-mail]>]

Fred Baker <[e-mail]> wrote:

> Simon, take a look in rtgwg. There is some work happening for source/destination routing in OSPF, IS-IS, BGP, and I believe Babel, and a problem statement being considered for adoption in rtgwg as we speak. The BGP prototype is operational in CERNET2, I'm told.
> 
> https://datatracker.ietf.org/doc/draft-baker-ipv6-isis-dst-src-routing
> https://datatracker.ietf.org/doc/draft-bowbakova-rtgwg-enterprise-pa-multihoming
> https://datatracker.ietf.org/doc/draft-ietf-ospf-ospfv3-lsa-extend
> https://datatracker.ietf.org/doc/draft-ietf-rtgwg-dst-src-routing
> https://datatracker.ietf.org/doc/draft-ietf-rtgwg-multihomed-prefix-lfa
> https://datatracker.ietf.org/doc/draft-xu-ospf-multi-homing-ipv6
> https://datatracker.ietf.org/doc/draft-xu-src-dst-bgp

I think you've missed the issue I'm looking to address - on a quick peruse, it looks like only the second of those links addresses source address selection by the host/service. While these various routing protocol extensions may well be important in the larger networks, I believe the vast majority of small to medium business have no need of those - the big problem is not how to route the packets for a given source address(prefix), but how to select the source address(prefix) in the first place.

Taking my own situation at work. I have a single router - it's actually a Debian 7 (Wheezy) virtual machine and already doing source based routing for IPv4 traffic across two egress routes. Extending that to IPv6 is relatively easy - "simply" a matter of configuring the routing rules in much the same way as is currently done for IPv4.
This is echoed at some of our customer sites where they have a "small business" grade router (typically Draytek) supporting multiple WAN interfaces. The Draytek routers have policy based routing - "traffic matching this set of criteria should be routed via that WAN interface".

Where there are multiple routers (eg separate routers provided by each ISP), then my *assumption* would be that the IPv6 stack would forward packets to the router which advertised the prefix containing the IP address the host/service has chosen to use for the outbound connection. In general we would frown upon this setup since it doesn't work well in the IPv4 world - in the general case it means manual configuration on nodes to select the egress router and is inflexible.