RE: TE Requirements Draft - ELSP

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

Hi all,

I agree with Nabil when it states that current requirements PRECLUDE folks who are E-LSP capable from utilizing DS-TE. In fact consider the following example, where the network supports 3 Class Types:
 - CT2=EF
 - CT1=AF1+AF2
 - CT3=BE
Also imagine that on a given link the bw constraints are P1=20Mb/s, P2=60Mb/s, P3=100Mb/s, and that at a given moment in time the link is occupied only by 18Mb/s of EF traffic. A new E-LSP carrying an EF+BE flow of 10Mb/s total traffic, divided into 5Mb/s of EF traffic and 5 Mb/s of BE traffic should be refused by the link (in fact it cannot accept more than 20Mb/s CT2=EF traffic).

My question is: how can the network realize that the flow must be refused (or routed elsewhere) if there is no way to signal that the EF bw is 5Mb/s? Since the current signaling only allow to signal the total bw (i.e. 10Mb/s), how can we decide to admit or to reject the incoming E-LSP?

There are two possible answers:

(1) use L-LSP and set up two different 5Mb/s LSPs, one for EF and one for BE. Right, but if I'm not interested in the added value of L-LSP (i.e. the possibility to perform per-class routing and per-class protection/restoration) why should I use it? I see no advantages and at least one drawback (i.e. an increased number of LSPs to manage and longer tables in the LSRs).

(2) add the standards with the possibility to signal per-PSC BW requirements. A possible proposal in this direction is given in:
http://www.ietf.org/internet-drafts/draft-iovanna-rsvp-mpls-flowspec-00.txt


Best regards
Roberto


> -----Original Message-----
> From: Nabil Seddigh [mailto:nseddigh@tropicnetworks.com]
> Sent: Friday, November 16, 2001 11:47 PM
> To: Francois Le Faucheur
> Cc: TE WG
> Subject: Re: TE Requirements Draft - ELSP
> 
> 
> Francois,
> 
> > Or are you suggesting operations of DS-TE over E-LSPs which 
> > transport multiple OAs? In that case, what are the Service 
> > Provider requesting this?
> 
> Yes to the former. You already heard from Jim Boyle wrt the latter.
> There are other service providers interested in this. I am not sure
> if they are willing to indicate their positions on the list. Jim gave
> an example application. There are other reasons folks are interested
> in using E-LSP with DS-TE. e.g. protection, l2vpns etc.
> 
> > Could these SPs clarify whether:
> > -  they would use this to do per-class admission control but not
> > do per-class routing (would it not be a pity to deploy all this
> > sophistication and have voice not being able to take its shortest 
> > path simply because there is no more resources for data on a link)?
> 
> We should be careful not to pre-suppose a solution in the 
> requirements.
> In some networks and service solutions it makes sense to route
> each class (or OA) on a different LSP while in others may want
> to route on the same LSP. In your statement above, you mentioned 
> that it would be a shame if voice cannot take the shortest path
> due to lack of resources for data. I would rephrase this to say
> that as long as constrained based routing algorithms (which are
> vendor specific) can select paths to meet the voice constraint,
> this should not be an issue. 
> 
> Best,
> Nabil Seddigh
>