RE: How to progress DSTE, particularly p -> TE-Class[i] = {CT, p}

["Francois Le Faucheur (flefauch)" <flefauch@cisco.com>]

Hello Jim,

I've edited the RDM and MAM I-Ds to reflect the proposal I posted
earlier on the TEWG list regarding relationship with Shared Mesh
Restoration. The only item left before we can post the drafts and go to
last call is the issue you raised below on how to handle Link Bundles
and FA-LSPs. What's your recommandation on that? May I follow approach
2?

Thanks

Francois

>> -----Original Message-----
>> From: Francois Le Faucheur (flefauch) 
>> Sent: 27 August 2003 14:54
>> To: Jim Boyle; te-wg@ops.ietf.org
>> Subject: RE: How to progress DSTE, particularly p -> 
>> TE-Class[i] = {CT, p}
>> 
>> 
>> Jim and all,
>> 
>> I favor approach 2). Specifically, this separate document 
>> would make the
>> necessary generalisation to cover Link Bundles and FA-LSPs.
>> One of the benefits I can see in that approach is that we can be more
>> explicit about exactly how DS-TE works in the context of Link Bundles
>> and in the context of FA-LSP. It would make it very clear 
>> exactly where
>> DS-TE has an impact on each of Link-Bundling and FA-LSP and 
>> how it will
>> work.
>> Also, it would allow us to address details which are specific to the
>> DS-TE-over-Bundle or DSTE-over-FA-LSP and would not be captured by
>> generalisation statements a la "p is mapped to TE-class[i]". For
>> example, the lsp-hierarchy draft defines how the Traffic Engineering
>> parameters of a FA-LSP are derived including the Unreserved Bandwdith
>> and the Max Reservable bandwidth. Where DS-TE runs over the 
>> FA-LSP, it
>> may be that the natural behaviour would no longer be to set 
>> all intial
>> Unreserved Bw values to the FA-LSP bandwidth, but rather to 
>> set to zero
>> all those that relate to TE-classes whose CT is different to 
>> the FA-LSP
>> CT. Similarly, there is a question as to how you select the BC values
>> over an FA-LSP (for example assuming RDM, perhaps it should 
>> be that, if
>> FA-LSP belongs to CTn, then BCm=<FA-LSP Bandwidth> if m<=n, 
>> and BCm=0 if
>> m>n ). These rules are open to discussion of course and they 
>> may not be
>> the right ones, I don't know yet, but there are just 
>> examples of things
>> which are specific to DSTE operations over FA-LSP and that 
>> we may want
>> to think about and specify.
>> Another benefit of course, is that we can finalise DS-TE 
>> proto without
>> any further delays.
>> 
>> If we were to go with approach 1), we would also need some words to
>> capture the other dimension of the generalisation, which is that
>> references to "Max Reservable Banwdith" should be mapped to "<Max
>> Reservable Bw, BCs>"
>> 
>> I agree 3) is not very viable.
>> 
>> Cheers
>> 
>> Francois
>> 
>> >> -----Original Message-----
>> >> From: Jim Boyle [mailto:jboyle@pdnets.com] 
>> >> Sent: 19 August 2003 07:03
>> >> To: te-wg@ops.ietf.org
>> >> Subject: How to progress DSTE, particularly p -> TE-Class[i] 
>> >> = {CT, p}
>> >> 
>> >> 
>> >> 
>> >> One issue that draft-sivabalan-diff-te-bundling-02.txt 
>> >> highlights is that 
>> >> there may be text in existing RFCs and I-Ds for MPLS/TE 
>> >> which address how 
>> >> to support priority preemption between LSPs.  
>> >> 
>> >> draft-ietf-tewg-diff-te-proto-04.txt redefines the 
>> >> reservable bandwidth 
>> >> available at priority "p" to be the reservable bandwidth 
>> >> available for 
>> >> TE-Class "i" where that is defined as:
>> >> 
>> >>      TE-Class[i]  <-->  < CTc , preemption p >  
>> >> 
>> >> Where this mapping is defined locally.
>> >> 
>> >> This creates a problem, as improvements to the base of MPLS, for
>> >> instance bundled links refer to how to address the 
>> >> particulars of that
>> >> improvement with regards to priority, would also likely beof use e
>> >> might want to in a diff-serv TE environment as well.
>> >> 
>> >> It is obvious that one should assume that where one reads 
>> >> about "p" they 
>> >> should be thinking TE-Class[i], however it would be best for 
>> >> this to be 
>> >> explicitly stated somewhere.
>> >> 
>> >> Before the DSTE proto draft and the BC models progress to WG 
>> >> last call, 
>> >> I'd like to see this resolved.
>> >> 
>> >> There are 3 approaches that I've heard suggested.
>> >> 
>> >> 1) Address this in the diff-te-proto draft. State that for 
>> >> the IGP and
>> >> RSVP RFCs, as well as technologies that improve upon them
>> >> (e.g. FA-LSP, link bundling, etc..), in order to be diff-serv TE
>> >> compliant, you need to map all references of "p" to TE-Class[i]
>> >> 
>> >> 2) Address this in a parallel document to diff-te-proto, and let 
>> >> diff-te-proto progress to IESG before this is resolved.  
>> >> This parallel 
>> >> document will somehow generalize "p" in base documents, and 
>> >> diff-te-proto 
>> >> will fit on-top of that (somehow)
>> >> 
>> >> 3) Generalize "p" in the base documents, RFC3209, the 
>> >> igp-traffic drafts, 
>> >> all of the other MPLS/TE and GMPLS drafts as well.  For the 
>> >> record, I am 
>> >> not personally in favor of this approach.
>> >> 
>> >> Please advise how you propose we move forward with this,
>> >> 
>> >> thanks,
>> >> 
>> >> Jim
>> >> 
>> >> 
>> >> 
>> 
>>