The MPLS WG Archive

["Choudhury, Sanjaya" <Sanjaya.Choudhury@marconi.com>]

> 	Hi! I Now there are 4 different documented approaches related 
> 	to support of DiffServ-TE by IGPs (lefaucheur,boyel,kompella,
> 	ash,bitar). As far as I know, the last round of discussion on this
> 	issue, ended without any consensus. I thought it may not be 
> 	bad idea to have another round of discussion.
> 
> 	In my opinion, we can broadly divide the discussions into
> 	two areas: 
> 		(a) Whether (and How) to advertise the per-class(/type)
> 		bandwidth information via IGPs
> 		(b) Pre-emption
> 
> 	To limit the scope of discussion, in this thread, let us discuss 
> 	the first issue only.
> 
> 	Here are some of my observations and thoughts (If my observations
> 	from existing drafts are incorrect please correct me).
> 	
> 	Your coments will be appreciated.
> 
> 	Thanks,
> 	sanjay
> 
> 	Q1. Do we need to advertise the per-class (/type) bandwidth
> 	information via IGP ?
> 	
> 	 -As far as I understand, two of the drafts  (ash,boyel?) prefer
> 	 not to advertise per-class bandwidth info for scalability reasons.
> 
> 	-In my opinion, it is better to advertise the per-class bandwidth
> 	info via IGP. This information, will help the path computation 
> 	module to do a better job. And one can handle, the scalability
> 	problems, via different mechanisms (as illustrated by other 
> 	drafts). 
> 		
> 	Q2. If we need to advertise the per-class(/type) bandwidth 
> 	information, Should the domain administrator have control
> 	over the granularity of the information get advertised in his
> 	domain. ?
> 
> 	-In my opinion, the domain administrator should have control 
> 	the per-class BW info that gets advertised in his domain via 
> 	a MIB (bandwidths for classes that are not advertised can be
> 	folded into the un-reserved bandwidth associated with the link)
> 
> 	Q3. How do we advertise the per-class(/type) bandwidth info?
> 		(a) 1-1 mapping with diffserv classes
> 		(b) m-1 mapping with diffserv classes (lefaucheur)
> 
> 	-In my opinion, 1-1 mapping is simpler and achieves the purpose
> 	without introducing new concepts (class-type).
> 	-1-1 mapping
> 		-one of the drafts (bitar) explicitly advertises the BW 
> 		associated with the class
> 
> 		-the other draft(kompella) overloads the diffserv-class
> 		with the priority information.
> 		
> 		-which approach is better ?
> 
> 	Thanks,
> 	sanjay
> 
> 	
> 	
> 	
> 	
> 	
>