Re: [OSPF] draft-chen-ospf-transition-to-ospfv3-00.txt

[Karsten Thomann <[e-mail]>]

Am Montag, 10. Februar 2014, 14:33:05 schrieb RJ Atkinson:
> Earlier, Karsten Thomann wrote, in part:
> > I don't know if i have missed that part, but what would happen if OSPFv3
> > is running over IPv4 and there are two (or more) IPv6 Islands already
> > deployed within the network, should there be any LSA flooding checks,
> > like do not flood LSA with IPv6 networks over IPv4 only links, or should
> > the route
> > calculation/installation of the information simply fail
> > as there is no valid path?
> 
> The OSPF deployments I am familiar with did not "auto deploy"
> or "self deploy".  Instead, all of the OSPF deployments that I am
> familiar with had engineers making deliberate decisions about how
> to configure the OSPF deployment, and also how to configure the
> OSPF routers in that deployment.
I know that such deployments are normally well planned, but as there is already a draft about 
ospfv3 autoconfig, even if it is not designed to be used in company networks.

> One deployment option, already deployed in some places, is to run
> "ships in the night".  In this model, the IPv4 topology exclusively
> uses OSPFv2 over IPv4 and the IPv6 topology exclusively uses OSPFv3
> over IPv6.  For some sites, that is a good option.  A number of my
> clients are very interested in this new I-D because it optimises
> their IPv6 transition strategy.
> 
> For some sites, IPv6 transition is simplified, optimised, and also
> cost-reduced by using OSPFv3 over IPv4 initially, and carrying both
> IPv4 prefixes and IPv6 prefixes inside that one OSPFv3 deployment
> (i.e. using the Address Family extension to keep the prefixes clearly
> differentiated within a single OSPFv3 deployment).  This can lower
> operational costs because one can get by with managing only OSPFv3,
> rather than having to manage both OSPFv2 and OSPFv3.
> 
> If one's "IPv6 islands" (to use your term) are really dual-stack
> (aside: having dual-stack rather than IPv6-only would be expected
> and normal if another part of one's OSPF deployment is using
> OSPF over IPv4), then EITHER the whole deployment should be configured
> to run over IPv4  XOR  the whole deployment should be configured
> to run over IPv6.
> 
> Having different "IPv4 islands" and "IPv6 islands" is an explicit choice
> of a configuration engineer or design engineer.  That probably is not
> the optimal deployment choice for an engineer to make.  It definitely
> is NOT a choice that I would make or that I would recommend to my clients.
I would not do it either, but I have already seen some strange ospf designs..., and as mentioned it 
is a case which can happen with ospf autoconfiguration, but we don't need to discuss it further as 
it is for homenetworks

> From the descriptions and definitions in the draft, it would not be
> "normal" or "expected" to mix IPv6 transport with IPv4 transport
> within a single OSPFv3 deployment.
> 
> > Is there any explicit preference over which version the adjacency
> > is build if v4 and v6 is available in that specific part of the network?
> 
> The choice of preferred OSPF transport really is something that the
> configuration engineer ought to configure.  As the details of configuring
> any IP router are outside the scope of the IETF, I am not sure that
> this I-D properly can or should say very more than it already says.
> 
> Perhaps the I-D could add clarifying text along these lines
> somewhere in Section 2:
> 
>  "Implementers of OSPFv3 over IPv4 SHOULD add a configuration
>   knob to let the router administrator select whether a given
>   OSPFv3-enabled interface should use OSPFv3 over IPv4 or
>   OSPFv3 over IPv6.  Except when an OSPFv3 deployment is being
>   transitioned from using IPv4 transmission to using IPv6
>   transmission, it is RECOMMENDED that all routers within an
>   OSPFv3 deployment use the same IP version for OSPFv3 packet
>   transmission."
> 
> > What should happen when an IPv4 Link gets v6 added, graceful reestablish
> > the adjacency over IPv6, or wait until forced protocol switch?
> 
> See my answer just above.
> 
> I would expect that any router that implements this I-D also would implement
> a configuration knob to indicate whether IPv4 transmission or IPv6
> transmission is preferred.  Over time, a configuration engineer might
> change this (e.g. manually change from IPv4 transmission to IPv6
> transmission AFTER an entire IPv4 network has been
> deployed-with/converted-to dual-stack IPv4 and IPv6).
> 
> Yours,
> 
> Ran Atkinson

Thanks for the detailed anwsers, I'm satisfied with the addition.

Kind regards,
Karsten
> 
> 
> _______________________________________________
> OSPF mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/ospf