(6bone) multihoming (was: Ingress filtering (was: asymmetric routing))

[Jan Oravec]

Hello,

> biggest problem is that the load is mainly on the ip6.{arpa,int} root
> servers. (but delegation of prefixes to many servers would mitigate
> this).
> 
> mad idea?
very mad, there is a bigger problem:

this would drastically slow-down packet processing... DNS lookup may take
too long... routers would require enough large cache for this, etc.

The better would be to develop new routing protocol... It's proved existence
of protocol which would require amortized time O(log N) for processing
single update and memory O(K) for routing record of some prefix, where N is
number of prefixes advertised and K is number of peers on router. The
AS-PATH entry is probably good only for routing-loop detection (which can be
solved other way), debugging and making BGP useless for multihoming of every
site. Only few of us does routing based on the middle part of it. (usually
when peer A advertises us a prefix with as-path A B ... X, we just care
about A and X, not about part between). Thus we require memory O(NK), which
is not a problem for several millions prefixes. About time analysis,
amortized O(log N) per update gives us initial time O(N log N) for
establishing session. For N -> several millions it gives us convergence time
a few seconds. For IPv6 and advertisements of zones /48 and less, log N is
at most 48 and I can imagine hardware solution to reduce time to O(N), thus
we will be able to have billion of multihomed prefixes in presence.

At the moment I do not have enough time to work on developing it, be cause I
have a lot of work with our project XS26, but in 3-4 months I will be able
to do so.

I will consult that with multi6 WG when I have something ready.


Best Regards,

-- 
Jan Oravec
project coordinator
XS26 - 'Access to IPv6'  
jan.oravec@xs26.net