The MPLS WG Archive

["Roberto Mameli (ERI)" <Roberto.Mameli@eri.ericsson.se>]

Sanjay and Abdul,

in my opinion there are a lot of possible scenarios in which E-LSP are useful... some of them are described in appendix A of draft-ietf-mpls-diff-ext-09.txt (see e.g. scenario 3).
Just to add another example consider a provider running only a couple of BAs (e.g. a premium and a standard class of service) that would like to traffic engineer its network. He could be simply interested in fair distribution of traffic at the purpose of avoiding congestion and increase efficiency, without the additional complexity of per-class routing and per-class protection/restoration functionality. In this case E-LSP is the right solution; it provides all the functionality requested in this scenario and also allows label space saving (as Abdul correctly states).

As far as per-PSC BW requirements signaling concerns, the standard currently doesn't cover this issue. In my opinion this represents an inconsistency. Just consider that the Class-Type proposed to reduce IGP flooding is not compatible with E-LSP (unless you find a way to signal BW per-PSC). A proposal in the direction of filling this gap is given in:
http://www.ietf.org/internet-drafts/draft-iovanna-rsvp-mpls-flowspec-00.txt

Hope it helps

Regards,
Roberto


> -----Original Message-----
> From: Abdul Malick [mailto:abdulmk@future.futsoft.com]
> Sent: Friday, November 16, 2001 3:31 PM
> To: Choudhury, Sanjaya; 'mpls@UU.NET'
> Subject: RE: How are the domain administrators are planning to use
> E-LSPs?
> 
> 	Hi! I have some basic questions related to the usage of E-LSPs.
> 	I will appreciate, if you can help me with the 
> following questions:
> 
> 	Q1. Is any domain administrator/service provider actually using
> 	(planning) to use the E-LSPs in their network ? In which cases,
> 	E-LSPs are more useful than L-LSPs?
> 
> 
> 	>> One advantage of using E-LSP, i think is the 
> conservation of Labels. In
> case we know that the path taken by an LSP is capable of 
> supporting all the
> required services. Then we can conserve labels by using one 
> E-LSP to support
> more than one PSC (to support the required services). In case 
> of L-LSP we
> will have to use one LSP for each PSC and hence waste more 
> labels. Is that
> right?
> 
> 	Q2. While deploying E-LSPs in the network, is the primary usage
> 	is limited to differentiation via the signaled DiffServ 
> classes ?
> 
> 	OR is there practical use for reserving (and enforcing) 
> bandwidth
> 	for the E-LSP  (per LSP / per PSC bandwidth specification) ?
> 
> 	>> I don't know if there is a means to signal the 
> distribution of the BW
> among different PSCs carried in an E-LSP. I think the present 
> approach is to
> use predefined distribution (configuration). Please correct me if i am
> wrong.
> 
> 
> 
> 
> 	Currently, the DS-TE-requirement draft does not address the
> 	E-LSP related requirements very clearly. Should it be updated
> 	with E-LSP related requirements?
> 
> 	Thanks,
> 	sanjay