Tom,
At 15:08 18/05/2010, Phelan, Tom wrote:
Hi Lars,
Well, I like option 1.
I feel that option 2 is a chimera. The 'G' in the GUT proposal stands
for "generic", but it is not entirely generic. The decapsulation stage
is specific to the encapsulated protocol. The GUT draft gives one decap
rule that is more-or-less suitable for TCP and DCCP. It doesn't work
for SCTP.
You have to look in detail at the encapsulated protocols to see if a
proposed decap works. With my non-detail understanding of TCP, it looks
like it works, but we'll need someone to look at that in detail.
The G in GUT is certainly not true for host implementations, but
importantly the wire protocol is generic for all e2e protocols it
encapsulates, which is what is important for middleboxes.
For DCCP, the proposed decap makes partial checksums ineffective. I
think it's OK to do that, but you need to state that explicitly in the
draft, and probably offer guidance on how to deal with partial checksum
feature negotiation. You also need to deal with how to signal the use
of UDP encap, and the other things included in my draft. Once you've
done that, you have my draft.
GUT has similar issues for SCTP. SCTP uses a CRC checksum that doesn't
include the IP addresses, so the proposed GUT decap doesn't work for it.
What's needed is actually easier to do. I don't know enough about SCTP
to know what else needs to be done.
Earlier on the tsvwg list I was proposing how to get GUT to do a
partial checksum that only covers its own headers, stopping at the
encapsulated headers.
As I've stated before, if you really think through a generic approach,
it becomes an overall scheme that can be used by all protocols, plus a
set of protocol-specific adaptations. That's where we're at with the
DCCP- and SCTP-specific drafts.
Tom P.
> -----Original Message-----
> From: tsv-area-bounces@xxxxxxxx [mailto:tsv-area-bounces@xxxxxxxx] On
> Behalf Of Lars Eggert
> Sent: Tuesday, May 18, 2010 3:38 AM
> To: tsvwg@xxxxxxxx
> Cc: DCCP working group; TSV Area
> Subject: Re: UDP encaps for SCTP and SCCP
>
> Hi,
>
> the discussion has touched on lots of things related to UDP encaps,
but I
> haven't seen anything I'd call consensus on the question below. I'd
> therefore like to ask folks to specifically state which option they
> support:
>
> (1) do one SCTP-specific and one DCCP-specific UDP encaps
> (2) do one generic UDP encaps that can be used with both
> (3) do neither (don't do any sort of UDP encaps for SCTP and DCCP)
>
> Thanks,
> Lars
>
> On 2010-4-22, at 12:57, Eggert Lars (Nokia-NRC/Espoo) wrote:
> > Hi,
> >
> > as most of you probably know, there are two different proposals for
how
> to encapsulate SCTP and DCCP inside UDP.
> >
> > One approach proposes two protocol-specific encapsulation schemes
> (described in draft-tuexen-sctp-udp-encaps and
draft-ietf-dccp-udpencap).
> >
> > The second approach proposes a generic encapsulation scheme that can
be
> applied to both SCTP and DCCP (draft-manner-tsvwg-gut).
> >
> > As a community, we do need to come to consensus on which of these
two
> approaches we want to follow when it comes to UDP encapsulation of
SCTP
> and DCCP. I believe it would be very confusing if we were to
standardize
> both approaches.
> >
> > I'd hence like to ask folks to read the three documents and post
their
> views to the tsvwg@xxxxxxxx list. I'm personally especially interested
in
> hearing from folks who aren't on the author lists of the documents,
but
> obviously, the authors expert opinions do matter.
> >
> > Thanks,
> > Lars
> >
> > PS: I'm pushing on this topic, because UDP encapsulation is the last
> remaining work item in the DCCP working group before it can close...
________________________________________________________________
Bob Briscoe, BT Innovate & Design
