Re: draft-van-beijnum-multi-mtu-05.txt

[David Lamparter <[e-mail]>]

On Thu, Apr 14, 2016 at 03:22:11PM +0200, Mikael Abrahamsson wrote:
> On Thu, 14 Apr 2016, David Lamparter wrote:
> 
> > That's actually the crux of it: this isn't Path MTU discovery.  The MTU 
> > discovered by this mechanism is the first-hop MTU for a lot of paths in 
> > case of a router and it's going to be distinct from the final PMTU 
> > towards a particular destination.  Plus, state on this only ever exists 
> > for direct neighbors, not for any random internet host.
> 
> I don't believe the draft says this.

It doesn't, but by tracking neighbors as pairs of (L2 address, IP
version) it behaves pretty much like that.

> What it does is discover the MTU to an IPv6 address on the same LAN. To 
> then infer that any packets sent VIA this node, is not something the draft 
> (at least explicitly) says. The case could be made to say that any packets 
> sent via the "discovered-to-support-larger-MTU-router" could use this 
> higher MTU, but that's something that isn't obvious, and I think it's up 
> for discussion.
> 
> So if I have this network:
> 
> H1 - R1 - R2 - Internet
> 
> If R1 announces in RA to the LAN where H1 is, that MTU is 1500, but H1 but 
> R1 also sends HintMTU of 9180, should the default route H1 installs 
> based on this RA and points to R1 also have this higher MTU? Maybe. Maybe 
> not.

As mentioned in my other mail: The RA's MTU is for the link, not for the
router.  There is no MTU value for the route or router itself;  and
since the RA MTU value doesn't provide any number and we're bumping the
"effective" interface MTU towards the router, that does pretty much
imply the route to R2/internet inherits that value.

> As I said in the session, I would like more fine granular MTU, so that I 
> can have separate MTU for separate entities.

Well - host-wise, doing PLMTUD as Ole suggest would probably work out
fine.  Need routers to make the problem interesting ;)

> For instance, if I know that most of my devices in my home support 9180, 
> but none of my Internet connections do (they only support 1500), then I 
> would like H1 to have this in its routing table:
> 
> Local /64 HintMTU 9180
> RIO with home /56 HintMTU 9180
> RIO with ::/0 with HintMTU 1500
> 
> Because of this, H1 wouldn't need to spend a lot of time trying to send 
> packets to the Internet with larger MTU than 1500.
>
> This would however require that we allow MTUs tied to not only PIO, but 
> also RIO. I still don't know what to do about the implicit default route 
> that one points to anyone one sees an RA from. Personally I consider this 
> design of the IPv6 protocol to be unfortunate.

Hrm... indeed an annoying problem...


-David