RE: /128 IPv6 prefixes in the wild?
I think you will learn a lot of /128s from IGP, but not from eBGP. I consider the "wild" to be the DFZ or similar type of network and in that case, you should not see advertisements for anything longer than a /48. This is not hard and fast, but please correct me if I'm wrong. - Brian J.
-----Original Message----- From: Mark Tinka [mailto:mtinka@globaltransit.net] Sent: Thursday, December 15, 2011 12:30 AM To: nanog@nanog.org Subject: Re: /128 IPv6 prefixs in the wild?
On Thursday, December 15, 2011 01:54:56 PM Glen Kent wrote:
In an IP/MPLS world, core routers in the service provider network learn the /32 loopback IPv4 addresses so that they can establish BGP/Targetted LDP sessions with those.
That's right - not sure how things would have been if 'draft-swallow-mpls-aggregate-fec-01' had gained some traction.
They then establish LSPs and VPN tunnels.
Indeed.
Since we dont have RSVP for IPv6 and LDP for IPv6 (not yet RFC) we cannot form MPLS tunnels in a pure IPv6 only network. GIven this, would v6 routers have large number of /128 prefixes?
What are the scenarios when IPv6 routers would learn a large number of /128 prefixes?
I suspect ISP's that choose to assign broadband customers /128 addresses because "they only ever need one address" may be a situation where you see rise given to this.
I would presume that most IPv6 prefixes that the routers have to install are less than /64, since the latter 64 is the host part. Is this correct?
This is certainly going to re-open some "wounds", but no, not all providers are assigning /64 to interfaces. Some (like us) are using longer prefix lengths such as /112 and /126.
But as for /128 prefix lengths, aside from the fact that Loopbacks will be floating around the network, whether you're using them to signal MPLS LSP's or setup iBGP sessions, you will see them with ISP's that assign them to customers and choose not to aggregate them at specific edge routers.
Cheers,
Mark.
On Thursday, December 15, 2011 09:56:04 PM Brian Johnson wrote:
I think you will learn a lot of /128s from IGP, but not from eBGP. I consider the "wild" to be the DFZ or similar type of network and in that case, you should not see advertisements for anything longer than a /48. This is not hard and fast, but please correct me if I'm wrong.
Ideally, yes. Good filtering (against your peers, customers and upstreams) will ensure you keep anything longer than a /48 out of your AS. However, do note that if you provide customer-induced automated blackhole routing (where customers attach an "evil" community to an "evil" host route and send that to you in an eBGP update because you expect it), that's one other way to see /128's (or more appropriately, something longer than a /48) across eBGP sessions with customers. Also, if customers multi-home to you and they want to be able to load share traffic across the various links between their network and yours, you may be inclined to allow them to send longer subnets that have a NO_EXPORT community attached to them so that load sharing occurs within your network for their inbound traffic, but de-aggregated routes do not flood the rest of the Internet. This is another way you could get "longer" routes into your network, with the benefit of not polluting the global Internet. Among other scenarios... :-). Mark.
participants (2)
-
Brian Johnson -
Mark Tinka