[c-nsp] IPv6: Getting started

[sthaug at nethelp.no]

> Q: IPv6 auto-configuration on p-t-p core links: Good or bad idea? I
> guess the downside is comparable to using RFC1918 addresses on Internet
> core router's interfaces, where e.g. traces break.

Why on earth would you want to use anything other than static IPv6
addresses for your p-t-p core links?

> Q: We run an MPLS VPN network, global routing only used for management.
> According to this document (which should also cover 12.2SX):
> 
> http://www.cisco.com/en/US/docs/ios/ipv6/configuration/guide/ip6-ov_mpls_6vpe.html#wp1054029
> 
> > 6VPE supports an MPLS IPv4-signaled core. An MPLS IPv6-signaled core
> > is not supported.
> 
> We have to keep an IPv4 core then? I guess the important part is to
> enable IPv6 for the users and services, but (excuse my french) it seems
> a little half-assed to not support an IPv6 core. Especially MPLS being
> what it is.

If you want IPv6 VPN you still need an IPv4 core, yes. Just another
missing feature (and it's missing from the Juniper side too...)

> Q: Should our initial "test phase" with IPv6 be accomplished using
> FE00::/7 local addresses (as per RFC 4193), or should we just aim at
> starting out with globally unique addresses? (I assume they're almost
> trivial to apply for and have allocated.)

Get your own IPv6 prefix and start your testing with that. It's easy.

> Q: We currently use IPv4 PI address space. Any point(s) in not applying
> for IPv6 PI address space? We have ~25 larger geographic sites and ~150
> smaller. We're logically strictly hierarchical, so there isn't much need
> for a lot of prefixes. We currently use about 1/3rd of the /16s in
> 10.0.0.0/8 network wide, mostly /24 subnets. A single /48 would thus
> easily fit our needs.

If you're a RIPE LIR then you can easily obtain IPv6 prefix(es) from
RIPE.

Steinar Haug, Nethelp consulting, sthaug at nethelp.no