[ppml] 2005-1 status

[Scott Leibrand]

On 01/24/06 at 6:49am -0000, Christopher Morrow <christopher.morrow at gmail.c...:

> I'd point out something here, that might not need pointing out
> explicitly, shim6 MAY only work if both sides of the conversation
> decide to use the protocol.

It's not news (to me at least) but it's worth discussing.  :)

> So, if your small office situation (dsl and cablemodem connected and
> shim6 enabled) starts/tries to do shim6 with a 'content provider' site
> that has PI space and doesn't know from shim6 (some content providers,
> akamai, use strange kernel tweaks/os-tweaks and likely won't support
> shim6 'features' early on) doesn't shim6 isn't going to provide the
> multihoming and failover and TE (perhaps) it should be providing.

I would maintain that shim6 *support* and shim6 *use* are somewhat
different things.  For example, suppose that new OS versions start
shipping with shim6 support enabled by default, but configured not to
actually use shim6 by default.  In that case, the default behavior would
be to never initiate shim6 compatibility negotiation, but to accept shim6
initiation requests from peers.  If a client is multihomed with multiple
addresses per host, and needs/wants shim6, they can change the defaults to
initiate shim6.

In addition, much of the traffic to large akamaized websites may not need
shim6 capabilities.  It's really only needed for longer-lived sessions.
Connectivity to the average website isn't really impeded by a change in
the client's IP address because the average TCP session is so short.

> I'd venture to guess it'll be a whole lot less useful than even it's
> proponents wish it would be, with the situation above in mind atleast
> :(

That may be so, but IMO that's because it's only needed in a minority of
cases, the ones where sessions are long-lived.  But I think that with
shim6 support, RFC3484 extensions for source address selection, and
implementation of simple source-based routing to avoid tripping strict
uRPF filters, multihoming can indeed be a possibility for end users and
small sites who can't afford to run BGP.  Call me naive, but I think that
once the protocols are finalized and end users start to get access to
IPv6, there will be enough demand for such functionality that it will be
included by default in IPv6 stacks going forward.

-Scott