Re: [dtn-interest] FW: [IRSG] POLL: draft-irtf-dtnrg-arch-07 -- DUE 30 Oct

[Scott Burleigh <[e-mail]>]

Henderson, Thomas R wrote:
> FYI, I provided the below comments in the review process of this
> document.
> 
> Tom
> 
> -----Original Message-----
> From: Henderson, Thomas R 
> Sent: Monday, October 30, 2006 1:20 PM
> To: 'Aaron Falk'; Internet Steering Group
> Subject: RE: [IRSG] POLL: draft-irtf-dtnrg-arch-07 -- DUE 30 Oct
> 
> Aaron, this is my first careful review of this document.  I would lean
> towards 'not-ready-to-publish' until some more clarity in the
> architecture and terminology is introduced:

Hi, Tom.  As with Michael's review, I think somebody who's closer to the text of the Architecture document than I am (Leigh or Kevin) should respond to most of your points.  I do have a couple of comments, in-line below.
 
> I did not find the terminology to be well defined.  In particular, the
> architecture introduces custom terminology (bundles, blocks, bundle
> fragments, custodians and custodial transfer) that take the place of
> more conventional networking terminology.  I recommend at the least that
> there is a terminology section up front that defines these things in
> relation to more conventional terms, or adoption of the more
> conventional terms.

I think a terminology section up front is a good idea.
  
> Also, I found the use of the term endpoint identifier (EID) to be
> non-standard; it seems like what you are really talking about are
> addresses, or maybe they are serving both functions, in which case it
> should be so stated.  The EID seems most equivalent to an IP address, or
> perhaps an i3 trigger since it is late binding.  I would suggest not to
> overload the term EID with this concept.  Also, is routing and
> forwarding done on EIDs?  Does each bundle (and bundle fragment) carry
> the EID?

My preference would be not to start all over again with terminology wrangling.  One thing I would like to clarify: bundle protocol EIDs are not addresses, because they do not necessarily have any topological significance at all (unlike the subnet number structure of IP addresses, the street number structure of postal addresses, etc.).  You route on them but nominally you do so only indirectly: EIDs are names that you use to look up or infer (somehow) the topological relationships you need to make routing decisions.

> It is not clear whether ADUs or bundles are passed to the DTN stack
> across the service access point (SAP).  In 3.1 and 3.3.1, it says that
> applications send ADUs, but in other places, it describes bundles being
> the unit of data passed.  But it says in 3.1 that ADUs may be fragmented
> into bundles; who does this fragmentation?

It's only ADUs that are passed across the service access point, and only ADUs are fragmented.  This may be a little more explicit in the Bundle Protocol spec.

Scott

> Section 2, it seems like an additional guiding principle for
> applications is that they should communicate data lifetimes to the
> network when possible.
> 
> In section 3.4, the sentence "Group join operations are initiated at
> receivers" seems out of place in this section talking about naming of
> groups.
> 
> The reference to section 3.11 within Section 3.7 should refer instead to
> 3.12.
> 
> In section 3.8, second paragraph, s/We assume C(t) and B(t)/We assume
> C(t) and D(t)/
> 
> It's not really clear how fragmentation works; is the protocol control
> information copied into the fragments?
> 
> In the acknowledgements, s/Andrea/Andrei
> 
> -Tom
> _______________________________________________
> dtn-interest mailing list
> [e-mail]
> http://mailman.dtnrg.org/mailman/listinfo/dtn-interest
>