Re: Address Family Support in OSPFv3

[Acee Lindem <[e-mail]>]

Vivek Dubey wrote:
> Sounds interesting, as IGP would then not depend
> on "address family combinations" supported at IP
> level.
> How will multiple OSPFv3 instance will provide that
> support?

Alternately, you could also use separate OSPFv3 instances
for each address family. OSPFv3 natively supports multiple
instances of OSPFv3 on a link without using secondary
addresses or multiplexing via a layer 2 encapsulation (e.g.,
802.1q).

For incongruent IPv6 topologies, you'd simply configure
additional OSPFv3 instances using the interfaces you wish
to use for that topology. The OSPFv3 instance would populate
the RIB corresponding to the respective topology. As example
application would be carrying multicast traffic on a separate
topology from unicast traffic.

For other address families (most notably IPv4), we'd still
need new LSAs to carry the non-IPv6 prefixes (independent of
whether you use multiple instances or support multiple address
families in a single instance).

The advantage of supporting multiple address families in
single instance is that you only have a single set of OSPFv3
adjacencies. The disadvantage is complexity we're exerting
on the base protocol. To it right, you really need separate
interface costs, prefix ranges, redistribution policy,
default origination, etc, for each address family. You
already have all this with separate instances.

At the last IETF, there was discussion as to whether or
not there is a strong enough requirement for this to warrent
consideration by the OSPF WG. Much of this discussion focused
on whether or not a provider would ever phase out OSPFv2 in
favor of using OSPFv3 to advertise IPv4 prefixes.

My feeling is that perhaps we should consider the modest
changes to the OSPFv3 protocol now in case we ever want
to support multiple address families.

This E-mail contains the additional considerations I eluded
to in my previous post.

Thanks,
Acee

>
>
>
> On Wed, 11 Jun 2003 Acee Lindem wrote :
>
>>At the last IETF, the draft draft-mirtorabi-ospfv3-AF-00.txt was
>>presented. It proposes to make some protocol changes to OSPFv3 now
>>in case we ever want to support multiple address families (e.g.,
>>permutations of IPv4, IPv6, unicast, and multicast).
>>
>>The draft raised a moderate level of technical discussion (mainly
>>centered on the proposal's backward compatibility mechanism). The
>>big question is whether or not there is a real requirement for this?
>>We all know this is done in ISIS but that doesn't necessarily mean
>>there is a requirement. And if there is, could the requirement better
>>be satisfied with multiple OSPFv3 instances. On the other hand, we
>>really want to get the protocol changes in as early as possible if
>>we ever want to support multiple address families.
>>
>>Comments? I have some additional considerations that I will put
>>in a separate E-mail.
>>
>>--
>>Acee
>>
>>
>> ------------------------------------------------------------------------
>>
>> <http://www.herohonda.com/karizma>
>


--
Acee