Re: [v6ops] new draft: draft-taylor-v6ops-fragdrop

[Lorenzo Colitti <[e-mail]>]

On Mon, Oct 29, 2012 at 5:08 AM, Mark Smith <[e-mail]>wrote:

> I don't really disagree with that either, it's more that fragmentation is
> a current capability of IPv6, so I think the IETF's recommendation should
> be that it is enabled by default on the Internet. I think that
> recommendation fits the robustness principle too - "..., be liberal in what
> you accept from others".
>
> If the IETF recommends against forwarding fragments, then I think
> that creates the obligation to provide advice and methods on what do
> instead,
> which might include specifying a standardised UDP fragmentation
> method, and provide alternative methods for protocols that utilise IP layer
> fragmentation.
>
> Fragments may still be in use, however they may not used much at the
>
> application layer. OSPFv2 and OSPFv3 uses them instead of
> implementing it's own fragmentation mechanism (and that may be across
> multiple hops in the case of a OSPF virtual link), and they are likely to
> be commonly used in IPsec and GRE tunnels (PMTUD across the tunnel path and
> having the tunnel MTU adjusted to it is a more advanced capability).


Ok, so how about stating the following, then?

1. Fragments are unreliable on the Internet today.
2. If you're an application developer and your apps need to run on the
Internet, then fragments will likely not work in many cases. If you want
your application to work in these cases, you should either use
application-layer fragmentation, use path MTU discovery, or limit yourself
to small packets instead of sending fragments.
3. If you control the network (e.g., if you're the network operator), and
want to use fragments for tunnels or other purposes, go ahead, but note
that in other people's networks, they might be filtered.

Your suggestion of providing a generic UDP segmentation mechanism is a fine
idea, and one that we discussed before this draft was written, but our
feeling was that such a solution would be impossible to agree on until we
have a problem statement. This draft ("FYI: some operators filter
fragments. Keep that in mind.") was supposed to be the problem statement.