RE: Meta-issues: On the deprecation of the fragmentation function

["Templin, Fred L" <[e-mail]>]

Hi Ron,

> -----Original Message-----
> From: Ronald Bonica [mailto:[e-mail]]
> Sent: Wednesday, July 10, 2013 10:12 AM
> To: Templin, Fred L; Doug Barton
> Cc: [e-mail]
> Subject: RE: Meta-issues: On the deprecation of the fragmentation
> function
> 
> 
> >
> > Sure, the tunnel ingress can probe the path to the egress; such a
> > probing method is already covered by SEAL.
> 
> Most GRE implementations do this, too.
> 
>  But, if the path MTU will
> > not accommodate a packet that after encapsulation is as large as
> > (1280 + HLEN) there is no alternative for the ingress other than to
> > start fragmenting since the ingress is not allowed to send a PTB
> > message reporting a size smaller than 1280.
> 
> I understand that you want to solve for the use-case in which a tunnel
> interior link has MTU < (1280 + HLEN).

Yes; for example, a 1280 MTU tunnel crossing another 1280 MTU
tunnel.

> But before solving for that use-
> case, we need to do a cost/benefit analysis.
> 
> We understand the cost of solving for this use-case. The task of
> reassembly is moved to the egress router. So, we need to make sure that
> the egress router is large enough to handle the task of reassembly and
> we need to make sure that its resources cannot be monopolized by a DoS
> attack. We also have to maintain our fragmentation capability.

I understand that. But a couple of points:

1) the egress will never be asked to reassemble more than (1500 + HLEN)
2) the ingress will feel the pain too and will be motivated to tune out
   the fragmentation
 
> Note that some of the cost is absorbed by the owner of the egress
> router. However, a portion of the cost is absorbed by the entire
> community, as they deal with the operation complexity associated with
> fragmentation.

The community can help to get rid of fragmentation. All they have
to do is configure a larger MTU on links that connect routers to
other routers. The MTU should be (1500 + HLEN) or larger.

> Now let's try to understand the benefit. Is there an installed base of
> IPv6-capable links with MTU < (1280 + HLEN) that carry traffic between
> tunnel endpoints? Is there a reason why someone might want to design a
> network this way?

Tunnels within tunnels, as one example. They are a common use case
in operational practice today.

> If we were to solve for this use-case, who would be the beneficiary?

The entire Internet would benefit, since all barriers to larger
MTUs would be removed.

> Possibly the party deploying the MTU-challenged link? Or, asked another
> way, would cost be assigned to the beneficiary?

Not sure I understood that. Everyone benefits, but the tunnel
endpoints doing the fragmentation have to do what is necessary
to maintain the tunnel in any event.

Thanks - Fred
[e-mail]

>                                                               Ron
> 
> 
> 
>