Re: [mpls] comment on draft-ietf-mpls-residence-time

[Lou Berger <[e-mail]>]

Greg,

Thank you for looking at this. See below.


On February 4, 2016 2:11:12 AM Gregory Mirsky <[e-mail]> 
wrote:

> Hi Lou,
> I've followed on your suggestion to investigate use of already available 
> objects. I’ve looked into LSP_ATTRIBUTES Object [RFC5420]. Per my analysis 
> we can consider two options:
> •         introduce RSO sub-objects with RTM_Capability sub-TLVs ordered as 
> defined in the current RTM draft, to be used in LSP_ATTRIBUTES Object;

Given 5420 is silent on TLV ordering, I think you need a single RTM 
Attributes TLV, which contains your sub objects carried as sub-tlvs.

I think the new TLV could be allowed in both LSP_ATTRIBUTES and 
LSP_REQUIRED_ATTRIBUTES objects to cover when RTM is optional/required e2e. 
Just LSP_ATTRIBUTES would be equivalent to what you currently define.

> •         use RRO Attributes sub-object that will be present in RRO Object 
> and define new RTM Capability Flag(s) from IANA Attribute Flags registry. 
> Rules defined in Section 7.3 RFC 5420 would apply.
> Number of Attribute flags is variable, thus the latter option is as 
> flexible as the former or current option documented in our draft.
>

I ruled out the second option in my earlier email as I generally consider 
RROs  a weak approach to supporting LSP control (since they are subject to 
all sorts of generally policy related processing and are 'at risk' in large 
messages.)

That said, the introduction of address to this object does required a note 
in the document on the policy implications, e.g., that processing at  
boundaries such as envisioned in rfc4208 may require rro-like policy 
processing. But this doesn't change my feeling that piggybacking control on 
RRO is generally a 'bad idea'.

Lou

> Greatly appreciate your consideration, comments, suggestions.
>
> 	Regards,
> 		Greg
>
>
> -----Original Message-----
> From: Lou Berger [mailto:[e-mail]]
> Sent: Sunday, January 31, 2016 10:38 PM
> To: [e-mail]; [e-mail]
> Subject: comment on draft-ietf-mpls-residence-time
>
> Hi,
> 	I see you propose defining a new RSVP object class that has application 
> specific scope.  As I'm sure you know the RSVP class-number space is really 
> small so it's best to avoid allocating new object classes whenever 
> possible.  I believe there is an existing object class that you can use for 
> your purposes -- the LSP_ATTRIBUTES object.
>
> Have you considered carrying RTM Hops in 1 (or three,  depending on if you 
> want sub-tlvs or not) new RTM Attribute TLV (or TLVs)?
>
> If not, I think it is worth exploring the viability of this approach or any 
> other approach that doesn't require the allocation of a new class-num.
>
> Lou
> (As contributor and TEAS co-chair)
>