Hi Peng, Satoru,

Thanks for your comments and suggestions. Actually, I think there are two
kinds of routing issues around here, and we should take appropriate planning
for our stateless 4v6 routing system.

The first one is what Peng has mentioned above, that de-aggregated IPv4 IGP
addresses blocks together   with embedded non-continuous port set would have
impact on IPv6 routing system. This situation has no relationship with
shared-mode or exclusive-mode, and I personally think we can reduce this
impact with careful IPv4 address planning.

Here are two possible ways:
-- Guarantee that one shared-IPv4 address would only be allocated within the
same BNG (say, BRAS, etc). Thus, non-continous port info will no longer be
needed to shown up in IPv6 IGP and all the packets with the same shared-IPv4
address will be destined to the same BNG.
-- Select the same IPv6 stateless prefix with native IPv6 prefix in a
certain area. For example, if 2001:c68:1122::/48 is the prefix for one MAN,
we can also use it for stateless IPv6 prefix , followed by shared IPv4
address/port set. In this way, de-aggregated IPv6 routings can be further
aggregated in the upper level (say, backbone network). However, in this way,
this IPv6 prefix is relatively long and it would be difficult to ensure the
whole stateless IPv6 prefix (including IPv4 address/port set) to be less
than /64 (easier for SLLAC).

The second one is to deal with the co-existence scenario for shared-mode and
exclusive-mode (or other differentiated situation). As suggested by Satoru
and Remi, it can either be accomplished by IPv4 routing in stateless GW, or
by mapping rule specification. Anyway, it still have to do some kind of
 lookup, no matter whether it is a routing table or other kinds of database.
But I also agree that the performance would not be a big problem to handle
these additional prefix-based rules.

Best wishes

Qiong Sun




2011/7/29 Peng Wu <[email protected]>

> Hi Satoru,
>
> >> However, if we introduce multiple domains with multiple IPv4 prefix
> pools ( the number of each domain is N1, N2, N3,... Nm) , then the entry
> number of IPv4 routing table will be N1+N2+N3+..+Nm. In current situation,
> Nm would not be a small number anymore. So maybe some more optimization work
> will still be needed to make it more "stateless" ?
> >
> >What do you mean 'the entry number of IPv4 routing table'?
> >Do you presume that your IGP routing has almost 300K routes as same number
> of current internet full routes?
>
> If I get Qiong correctly, she means that, suppose the IPv4 IGP addresses
> blocks are quite de-aggregated already. If the ISP upgrade the network to
> IPv6, and support IPv4 using stateless 4v6 with the same IPv4 address
> distribution, then the IPv4 IGP kind of introduce the IPv4 IGP routing table
> into IPv6.
>
> If this situation does exists, then it's an interesting operation problem.
>
>
_______________________________________________
Softwires mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/softwires

Reply via email to