Re: [trill] Alvaro Retana's Discuss on draft-ietf-trill-channel-tunnel-10: (with DISCUSS and COMMENT)

[Donald Eastlake <[e-mail]>]

Hi Alvaro,

On Thu, Jul 7, 2016 at 6:51 AM, Alvaro Retana (aretana)
<[e-mail]> wrote:
>
> On 7/7/16, 1:16 AM, "Donald Eastlake" <[e-mail]> wrote:
>
> Donald:
>
> Hi!
>
> >>----------------------------------------------------------------------
> >> DISCUSS:
> >> ----------------------------------------------------------------------
> >>
> >> Even though the IANA Considerations section was just updated (in version
> >> -10), I am putting in this DISCUSS because it is still
> >> incomplete/incorrect.
> >>
> >> 1. Guidance for managing the SubERR namespace should be included.  Note
> >> that this document only specifies values for ERR 6, but guidance should
> >> be given to IANA for the other ERR values as well.
> >
> >No IANA registry is created for any SubERR values so I'm not sure why
> >there needs to be IANA guidance. The draft can be made clearer that
> >for all values of ERR other than 6, SubErr should be sent as zero and
> >ignored on receipt. If you want, a registry could be added. I would
> >think IETF Review was the appropriate registration procedure.
>
> Yes, I think a registry should be added.

OK.

> >> 2. Section 6.2.1 (RBridge Channel Error Codes Subregistry) requests the
> >> creation of a new registry ("RBridge Channel Error Codes²), but that
> >> registry was already created by RFC7178.  This document should then
> >>split
> >> the requests in two parts: assignment of the vales 6-8, and the change
> >>to
> >> the registration procedure.
> >
> >You are correct. Sorry for the oversight.
> >
> >However, I actually think the registration procedure should stay as
> >Standards Action for new ERR codes as provided in the current IANA
> >registry.New major error codes can cause substantial problems with
> >older implementations that do not know what they mean. (I do not think
> >this is a problem with SubErr since implementations can take a default
> >action based on the major ERR code.)
>
> That works for me; I just mentioned the registration procedure because it
> had been changed.

OK.

> >> ----------------------------------------------------------------------
> >> COMMENT:
> >> ----------------------------------------------------------------------
> >>
> >> >From Section 2. (RBridge Channel Header Extension Format), is the RESV4
> >> field a space that is reserved for potential future use?  Why isn¹t it
> >> ignored on receipt (similar to the RESV field in Section 4.3)?  If there
> >> is potential for use of this space (RESV is defined as 4 bits, which
> >> makes me think about potential bit-level allocations), then there should
> >> be some guidance in the IANA Considerations.
> >
> >Earlier versions of this document added three features to RBridge
> >Channel. As well as the payload feature and security feature in the
> >current document, which are invoked by non-zero values of the PType
> >and SType nibbles,
> >     if you look at the -01 or -00 version you will see a "Scope"
> >feature invoked by non-zero values of the Scope nibble. RBridge
> >Channel supports messages between RBridges in the same TRILL campus
> >and between an end station and an RBridge on the same link. The Scope
> >feature extended this to support, for example, messages between an end
> >station and an RBridge not on the same link. However, the scope
> >feature added significant complexity and there was no immediate
> >requirement for it so it was dropped and the Scope nibble was
> >relabeled RESV4.
> >    Those bits are deliberately made "critical" so the Scope feature
> >could be revived in the future and Scope-ignorant RBridges would know
> >to reject Extended RBridge Channel messages that are using the Scope
> >feature which changes the format a little. Any attempt to re-instate
> >the Scope feature or to make other use of these four bits will require
> >a Standards Track document in any case. I don't understand why there
> >needs to be any guidance in the IANA Considerations about this. What
> >would it say other than that any non-zero value is treated as an
> >error? which is a technical specification already present in the body
> >of the document and not an IANA Consideration.
>
> I agree with you on the IANA part (no need to add anything related to
> RESV/RESV4).

OK.

> I still think (non blocking comment) that treating a received non-zero
> value as an error adds unnecessary complexity if it can just be ignored.
> If the Scope feature is ever revived (or any other feature that wants to
> use the field) then it can change the behavior then.

Well, perhaps we can agree to disagree.
     If that nibble is made critical, it provides a strong guarantee
against mishandling packets of a somewhat different format indicated
by the nibble being non-zero even in bizarre cases of delay in
propagating link state indicating non-support if the configuration of
an RBridge is rolled-back or who knows what. On the other hand, if the
nibble is made "send as zero ignore on receipt", which I agree is
normally the right thing, you always have a small chance of something
slipping through and being misinterpreted and it gets complex to
decide what to do about cases where you have multi-destination RBridge
Channel messages. Say you have a target multi-cast group with 100
members and 1 is not advertising support of the Scope feature. It
might make sense to just send with the scope feature anyway, if that
is safe; 99 will process it and 1 discard it. But if the Scope nibble
is non-critical, you have to serially unicast 99 copies of the message
to be safe; ugh...
     I think a key thing here is that use of the Scope feature
requires the addition of a couple more address/label fields to the
packet and, without criticality of the nibble, those fields will be
misinterpreted by recipients. Of course, there are potentially other
ways around this by multiplying "protocols" and using different values
in other fields on which the recipient dispatches... Ultimately it is
a judgement call which is "simpler".

Thanks,
Donald
===============================
 Donald E. Eastlake 3rd   [phone] (cell)
 155 Beaver Street, Milford, MA 01757 USA
 [e-mail]

> Thanks!
>
> Alvaro.