Misconceptions, was: IPv6 RA vs DHCPv6 - The chosen one?

Masataka Ohta mohta at necom830.hpcl.titech.ac.jp
Fri Dec 30 09:49:48 UTC 2011


Ray Soucy wrote:

> But that is only the case if you let customers have a PI prefix (which
> I think is really required in a purist end-to-end model, but for the
> sake of argument...).

Multihoming by routing, by the intermediate systems, is against
the end to end principle, which is why it does not scale.

> The remote host would
> have no knowledge of other available prefixes

As IP layer is connectionless, let transport or application
layer carry the information on the prefixes to try to keep
connections.

 > (even if there is a path
> change, and a different path becomes favorable).

The remote host can use IGP metric to know the initial best
candidate and subsequent path changes.

It can be assumed that default free routing table is small.

In addition, the remote host may also use transport/application
layer timeout to try other prefixes.

> The result is still a very large amount of overhead. You will still
> experience significant change propagation delay (slower than change
> propagation under the current model

Not at all.

Transport/application timeout is no different.

Route propagation is no slower. Instead, smaller default free
routing table means more rapid convergence.

> This all would be significantly more complex than IPv6

It was IPv6 which was expected to support multiple addresses
to suppress routing table growth. The result is a complex
protocol with halfhearted support for multiple addresses.

> We now know that it takes well over a decade for
> people to move to a protocol, even when it is almost operationally
> identical to its predecessor.

Unlike IPv4, IPv6 is poorly operational and still needs protocol
modifications,

For example, multicast PMTUD causes ICMP implosion, which means
rational operators filter ICMP packet too big often including
those against unicast packets, which means PMTUD won't work.

Yes, fixing it requires more than a decade.

Then, it may be a good idea to obsolete SLAAC, which means
IPv6 address can be only 8B long. You know, remembering 16B
addresses is divine, which is an operational head ache.

> To be frank, you would have to build a completely new and
> parallel network and hope you could somehow convince people to adopt
> it

Multihoming with multiple addresses works at transport/application
layer over existing IPv4 and IPv6.

						Masataka Ohta




More information about the NANOG mailing list