Bjørn Mork bjorn at mork.no
Wed May 4 15:18:32 CEST 2011

Jared Mauch <jared at puck.nether.net> writes:
> On May 4, 2011, at 4:17 AM, Bjørn Mork wrote:
>> Jared Mauch <jared at puck.nether.net> writes:
>>> I'm not sure I would purchase colocation from anyone today that was
>>> unable to provide IPv6 on the same lan, even if it's some (ick) 6PE or
>>> (double-ick) tunnel.
>> What's the problem with 6PE? Whether the MPLS core is set up over IPv4
>> or IPv6 should not matter.   Or am I missing something?
> [dual stack model]
> If MPLS fails in my network, your IPv4 packets will still work.
> If MPLS fails in my network, your IPv6 packets will still work.
> [6pe model]
> If MPLS fails in my network, your IPv4 packets will likely still work

Not bloody likely.  If MPLS fails in my network, then the network is

> If MPLS fails in my network, your IPv6 packets may not make it
> anywhere, as there may be no *tunnel* for your traffic to reach the
> IPv6 enabled internet.
> -- snip --
> my issue is mostly regarding the ease of understanding the network
> architecture and excessive tunneling models.  (might as well break out
> the 3ffe space? ...)

well, for many of us MPLS is the basis for the current network
architecture regardless of protocol transported over the network.  It is
what's transporting the IPv4 packets around.  It is not some new magic
used only to transport IPv6. 

6PE is just a matter of making the existing PEs dual stack.  And the
beauty is that you can do this node by node, without even touching the
core routers.

> We are all overlay networks, be it over some IEEE, ITU or other
> transport, and ultimately an overlay network over the layer-1 fibers
> in the ground.  Each time you create an opaque tunnel you make it
> harder to diagnose what is wrong.  Trying to figure out who is
> dropping your GRE or IPIP or other traffic can be quite hard.  While
> MPLS is a tool, and it can be used to enable a variety of interesting
> services, they all have varying costs and consequences.  I'm of the
> belief that 6PE is a tool that should be short-lived in the same way
> as any 6to4, 4to6, CGN and similar services.

I agree that the "6PE" designation might die (in fact I'm not quite sure
why it was needed at all), but I'm pretty sure that IPv4/IPv6 dual stack
or IPv6 only PE routers will continue to exist for the forseeable future
(i.e. at least 24 months :-)

> Without E2E IPv6 we will be left in a situation where the IPv6 world
> will be adversely impacted by a bad IPv4 event.  The same can be said
> about providing IPv6 routes in IPv4 transport TCP sessions.  While you
> can mix them together, it's certainly less than ideal and I personally
> disagree with that level of fate sharing.

The routes should of course eventually be transported over IPv6 TCP
sessions, including the leftover IPv4 routes.  But there's still some
time until that will happen.  Until then it makes most sense to use IPv4
for this purpose.

Although I partly buy your argument about fate sharing, I don't think it
makes any sense duplicating network configuration by running dual stack
on every interface on every core router.  The additional configuration
is guaranteed to cause new errors, and the protocol losing is going to
be the least used one.  

By using 6PE you pretty much ensure that your IPv6 network is just as
stable as your IPv4 network is.


More information about the ipv6-ops mailing list