RE: Clarification on intent of requirement

I agree with what Caitlin said, with one further comment. This
requirement is really there to require the LLP to keep a message length
(for a byte stream like TCP) or have datagram semantics (for an LLP like
SCTP) such that a DDP segment can be pulled out intact at the opposite
end. 

I don't see that this talks in any way to an indication mechanism. This
is outside the scope of a wire protocol.




Jim



> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of
> Caitlin Bestler
> Sent: Monday, November 17, 2003 7:26 AM
> To: Barry Reinhold
> Cc: RDDP
> Subject: Re: [rddp] Clarification on intent of requirement
> 
> 
> On Nov 17, 2003, at 9:01 AM, Barry Reinhold wrote:
> 
> > Clause 5 item 4 of draft-ietf-rddp-ddp-01 states:
> >
> >    4.  The LLP MUST preserve DDP Segment and Message boundaries at
the
> >        Data Sink.
> >
> > For the LLP to preserve the DDP segment boundary I would expect that
> > the
> > LLP (MPA/SCTP) would, in some virtual API, provides an indication
that
> > the DDP segment is available (i.e. the full DDP segment is available
in
> > some buffer and is without error). Given an API this could be
tested.
> >
> > I am not clear on what the LLP is supposed to do with a DDP message
> > boundary. A DDP message boundary is defined by a bit in a DDP header
> > and
> > the semantics associated with processing a DDP message do not seem
to
> > be
> > appropriate for the LLP. In what way is the LLP supposed to
"preserve"
> > a
> > DDP message?
> >
> >
> 
> The requirement that the LLP "preserve Message boundaries" is perhaps
> misleading, because it is merely preserving the portion of the segment
> that indicates it is a message boundary. But the LLP is preserving the
> entire message, with no understanding of its internal structure, so of
> course it preserve whatever portion is responsible for marking message
> boundaries. If the draft authors want to clean the language up, fine,
> but it certainly is not an error as written.
> 
> Preserving Segment boundaries is more important, and what I understood
> the primary purpose of the clause to be. It is a re-affirmation that
> the service offered by the LLP is reliable delivery of DDP Segments,
> not byte streams.
> 
> Also, keep in mind that the virtual API between the LLP and DDP is
> allowed to have *two* methods: one for placement and a second for
> completion. At the time LLP completion is reported, the information
> on where the DDP Segment boundaries were must not have been lost.
> In particular the fact that TCP has acked a *portion* of a DDP Segment
> has no meaning to the DDP layer. It is not LLP completed until the
> entire segment has been LLP acked.
> 
> For the SCTP mapping, each DDP Segment is a single-chunk SCTP User
> Message, so there are no potential conflicts.
> 
> 
> _______________________________________________
> rddp mailing list
> [email protected]
> https://www1.ietf.org/mailman/listinfo/rddp