Re: [CCAMP] AD review of draft-ietf-ccamp-assoc-ext

Adrian,
	Thank you for the review.  Please see below for in-line responses.

Lou (as co-author)

On 8/8/2012 3:19 AM, Adrian Farrel wrote:
> Hi,
> 
> I have done my usual AD review of this document prior to issuing IETF
> last call. The purpose of my review is to catch any issues that might 
> come up in last call or IESG evaluation, and so smooth the path of the
> document.
> 
> This seems to be a well-written I-D, thanks. I have just a few small
> points that are mainly stylistic. A quick respin should address them.
> I start with one fundamental question that I hope can be answered by
> email.
> 
> Thanks,
> Adrian
> 
> ===
> 
> To start with, a key question:
> 
> Why does the working group want to publish this as an RFC when there
> is no immediate intention to implement for any of the many scenarios
> described in the document?
> 
> Will this become just another Standards Track RFC that gathers dust
> and obfuscates the real protocol specs, or is there good reason to 
> publish it?
> 

The document was motivated to support applications such as defined in
draft-ietf-ccamp-rsvp-resource-sharing (whose authors say they intend to
revive this work now that the foundation work has reach maturity.)

> ---
> 
> I can see how this updates 4872.
> 
> I am not convinced about this being an update to 2205, 3209 and 3473.

only as much as it provides additional capability to those RFCs, not
that it modifies their basic behavior. The point of listing them was to
make it clear that this draft's scope is beyond rfc4872 / recovery.
Perhaps it's sufficient to put this in the narrative and not as part of
the updates field.

> 
> Let me ask the question as follows: Is this document needed in order to
> produce a conformant implementation of RFC 2205 RSVP? 

No.

> In other words: is
> it your intention that all new implementations of RSVP MUST include an
> implementation of this document?

Absolutely not the intent.
> 
> If the document does make the updates stated, it would be good if some
> part of the document (maybe a new section, maybe just the Introduction)
> contained text that described what has been updated.

The relevant text is in section 3.  If this isn't sufficient can you
propose some alternate/additional language?

   The remainder of this section defines the general rules to be
   followed when processing ASSOCIATION objects.  Object usage in both
   Path and Resv messages is discussed.  The usage applies equally to
   GMPLS LSPs [RFC3473], MPLS LSPs [RFC3209] and non-LSP RSVP sessions
   [RFC2205], [RFC2207], [RFC3175] and [RFC4860].

> 
> Does Section 6.2 imply an update to RFC 4873?

Perhaps, it extends the definition provided in 4873 to non-recovery
application. implementations of rfc4873 aren't impacted by 3.3/6.2.  So
for this reference, I don't think so.

That said, 4873 does make use of the association object defined in 4872.
 4873 implementations can certainly make use of the Extended ASSOCIATION
Objects (section 4), but we didn't list an update to 4873 as 4873
references 4872 for the definition of ASSOCIATION Objects. So we thought
the 4872 update indication was sufficient.

> 
> ---
>                                         
> Section 1
> 
> Some ambiguity, since it looks like the extension is to recovery 
> contexts other than GMPLS!
> 
> OLD
>    This document expands the possible usage of the ASSOCIATION object to
>    non-GMPLS recovery contexts.  
> NEW
>    This document expands the possible usage of the ASSOCIATION object to
>    contexts other than GMPLS recovery.  
> END

I agree -- looking at it again after all this time.

How about (in multiple places)
OLD
non-GMPLS recovery
NEW
non-GMPLS and non-recovery

> 
> ---
> 
> Section 1 Voice Call Waiting
> 
>        However,
>        there is no way in RSVP today to share the resources between the
>        A->B and A->C subflows of the call since by definition the RSVP
>        reservations for these subflows must have different IP addresses
>        in the SESSION objects.
> 
> Since you are defining such a mechanism, "there is no way today" is not
> true. I suggest...
> 
>        Since, by definition, the RSVP reservations for the subflows A->B
>        and A->C of the call must have different IP addresses in the
>        SESSION objects, this document defines a new mechanism to
>        associate the subflows and allow them to share resources.

looks good to me. (assuming my co-authors don't object.)

> 
> Similarly...
> 
> OLD
>      o Voice Shared Line:
>        A single number that rings multiple endpoints (which may be
>        geographically diverse), such as phone lines on a manager's desk
>        and their assistant.  A VoIP system that models these calls as
>        multiple P2P unicast pre-ring reservations would result in
>        significantly over-counting bandwidth on shared links, since
>        today unicast reservations to different endpoints cannot share
>        bandwidth.
> NEW
>      o Voice Shared Line:
>        A voice shared line is a single number that rings multiple 
>        endpoints (which may be geographically diverse), such as phone
>        lines to a manager's desk and to their assistant.  A VoIP system
>        that models these calls as multiple P2P unicast pre-ring 
>        reservations would result in significantly over-counting 
>        bandwidth on shared links, since RSVP unicast reservations to
>        different endpoints cannot share bandwidth.  So a new mechanism
>        is defined in this document allowing separate unicast
>        reservations to be associated and share resources.
> END

okay (assuming my co-authors don't object.)

> 
> And...
> 
> OLD
>      o Symmetric NAT:
>        RSVP permits sharing of resources between multiple flows
>        addressed to the same destination D, even from different senders
>        S1 and S2.  However, if D is behind a NAT operating in symmetric
>        mode [RFC5389], it is possible that the destination port of the
>        flows S1->D and S2->D may be different outside the NAT.  In this
>        case, these flows cannot share resources using RSVP today, since
>        the SESSION objects for these two flows outside the NAT would
>        have different ports.
> NEW
>      o Symmetric NAT:
>        RSVP permits sharing of resources between multiple flows
>        addressed to the same destination D, even from different senders
>        S1 and S2.  However, if D is behind a NAT operating in symmetric
>        mode [RFC5389], it is possible that the destination port of the
>        flows S1->D and S2->D may be different outside the NAT.  In this
>        case, these flows cannot share resources using RSVP, since the
>        SESSION objects for these two flows outside the NAT have
>        different ports.  This document defines a new mechanisms to 
>        associate these flows and allow them to share resources.
> END

okay (assuming my co-authors don't object.)

> 
> ---
> 
> Globally...
> 
> s/extended ASSOCIATION/Extended ASSOCIATION/

sure

> 
> ---
> 
> Section 1
> 
> OLD
>    This document also defines the extended ASSOCIATION objects which can
>    be used in the context of Transport Profile of Multiprotocol Label
>    Switching (MPLS-TP).  Although, the scope of the extended ASSOCIATION
>    objects is not limited to MPLS-TP.
> NEW
>    This document also defines the Extended ASSOCIATION objects which can
>    be used in the context of the Transport Profile of Multiprotocol
>    Label Switching (MPLS-TP).  The scope of the Extended ASSOCIATION 
>    objects is not limited to MPLS-TP.
> END
> 

sure.

> ---
> 
> As with the previous ambiguity...
> 
> OLD
> 3. Non-GMPLS Recovery Usage
> NEW
> 3. Uses other Than GMPLS Recovery
> END
> 

propose same change as above: Non-GMPLS and Non-Recovery Usage

> ---
> 
> Section 3.1
> 
> s/defined generic/defines generic/

thanks.

> 
> ---
> 
> Section 3.1.2
> 
>    A node that wishes to
>    allow downstream nodes to associate Path state across RSVP sessions
>    MUST include an ASSOCIATION object in the outgoing Path messages
>    corresponding to the RSVP sessions to be associated.
> 
> Yuck!
> 
> Are the sessions to be associated in the outgoing Path messages or the
> Association object, or both. Actually, I don't think this needs a MUST,
> but can be converted to descriptive text as follows.
> 

Agreed. This sentence is clearly subject to misinterpretation and must
be fixed!

>    A node that wishes to allow downstream nodes to associate Path state
>    across RSVP sessions sends a Path message corresponding to one RSVP
>    session and includes in that message an ASSOCIATION object that 
>    indicates the associated RSVP sessions.

I think this actually doesn't match our intent.

See below for proposed wording as I think it's easier to parse the
sentences changes/together.

> 
>           
> Then you have...
> 
>    In the absence
>    of Association Type-specific rules for identifying association, the
>    included ASSOCIATION objects MUST be identical across all associated
>    sessions.
> 
> I know what you mean, but what you have said would require that a Path
> message must contain an Association object that refers to itself. I 
> think what you need has to be more verbose to be accurate.
> 
>    In the absence
>    of Association Type-specific rules for identifying association, each
>    Path message for a session in a set of associated sessions MUST
>    include an ASSOCIATION object that indicates all the other sessions 
>    in the set.

how about:
   To indicate association between multiple sessions, an appropriate
   ASSOCIATION object MUST be included in the outgoing
   Path messages corresponding to each of the associated sessions.
   In the absence of Association Type-specific rules for identifying
   association, the included ASSOCIATION object MUST be identical.

> 
> ---
> 
> 3.1.1 vs 3.2.1
> 
> In 3.1.1 you have...
>    Relative ordering of ASSOCIATION objects of the same
>    type SHOULD be preserved by transit nodes.
> 

note this is path processing,

> In 3.2.1 you have...
>    Relative ordering of ASSOCIATION objects of the same
>    type MUST be preserved by transit nodes.  Association type specific
>    ordering requirements MAY be defined in the future.
> 
Note this is resv processing.

> 1. Why is 3.1.1 SHOULD and 3.1.2 MUST?

I'm not sure, other than 3.1.2 is new so there's no chance of placing a
new requirement on implementations of 4872&3 (which only use association
objects in Path messages.)

I'm fine with should for both.

Co-authors?

> 2. Why does 3.2.1 consider association type-specific rules, but these
>    are not in 3.1.1?

3.1.1 relates to path messages and we didn't want to impose new
requirements on implementations of 4872&3.

> 3. s/type specific/Type-specific/

sure.

> 4. Why "MAY" not "may"?

over zealousness.  i.e., your right!

> 
> ---
> 
> 3.2.2
> 
> s/This section apply/This section applies/

Thanks.

> 
> ---
> 
> 3.3.1
> 
>    Once an association is identified, resources SHOULD be shared across
>    the identified sessions.
> 
> This use of "SHOULD" begs the question: under what circumstances MAY
> resource sharing not be applied?

actual resource allocation is really an implementation choice and
different internal implementation choices.  We didn't want to overly
constrain an implementation. "is expected" is really what we mean, but
how do you say this in 2119 language?

> 
> ---
> 
> Globally...
> 
> Please be consistent with:
> - Association ID
> - association-id

okay.

> 
> ---
> 
> Section 4.1
> 
> Global Association Source: 4 bytes
> 
>       This field contains a value that is a unique global identifier.
>       This field MAY contain the 2-octet or 4-octet value of the
>       provider's Autonomous System Number (ASN).  It is expected that
>       the global identifier will be derived from the globally unique ASN
>       of the autonomous system hosting the Association Source.  The
>       special value of zero (0) indicates that no global identifier is
>       present. Note that a Global Association Source of zero SHOULD be
>       limited to entities contained within a single operator.
> 
> Given the statement that the content is globally unique, I don't think
> there is room for you to pontificate about how this field might be 
> filled. You need to make a definitive statement. Otherwise the content
> cannot be guaranteed to be globally unique (unless you envisage a 
> central registration system!).

Humm this text is actually lifted from 6370 which says:

   Quoting from RFC 5003, Section 3.2:
      The global ID can contain the 2-octet or 4-octet value of the
      provider's Autonomous System Number (ASN).  It is expected that
      the global ID will be derived from the globally unique ASN of the
      autonomous system hosting the PEs containing the actual AIIs.  The
      presence of a global ID based on the operator's ASN ensures that
      the AII will be globally unique.

How about:

    This field contains a value that is a unique global identifier.
    This field MAY contain the Global_ID as defined in [RFC6370].
    The special value of zero (0) indicates that no global identifier
    is present. Note that a Global Association Source of zero SHOULD be
    limited to entities contained within a single operator.

> 
> ---                                                          
> 
> Section 4.1
> 
>    Extended Association ID: variable, 4-byte aligned
> 
>       This field contains data that is additional information to support
>       unique identification.  The length and contents of this field is
>       determined by the Association Source.  This field MAY be omitted,
>       i.e., have a zero length.  This field MUST be padded with zeros
>       (0s) to ensure 32-bit alignment.
> 
> This is confusing. I think that the length of the field is chosen by the
> Association source and determined by the Length field of the Association
> object. 
> 
> But your text implies that the contents of the field may be different 
> from a four octet quantity. Since you have not included a separate 
> length indicator, this cannot be. The object length must always be a
> multiple of 4 (says RFC 2205) and so there is no difference between a
> zero padded field and a field where the zeros have meaning.
> 
> You *could* say that the length of this field is Association Type-
> specific, but that would be ugly.

How about:
   Extended Association ID: variable, 4-byte aligned

      This field contains data that is additional information to support
      unique identification.  The length and contents of this field is
      determined by the Association Source.  The length of this field
      is derive from the object Length field and as such MUST have a
      zero length or be 4-byte aligned.  A zero length indicates that
      this field is omitted.

> 
> ---
> 
> 
> While you're at it, please note that draft-ietf-ccamp-assoc-info has
> been published as RFC 6689

Sure.

Thanks for the comments.

> 
> _______________________________________________
> CCAMP mailing list
> CCAMP@ietf.org
> https://www.ietf.org/mailman/listinfo/ccamp
> 
> 
> 
> 