The MPLS WG Archive

[Nabil Seddigh <nseddigh@tropicnetworks.com>]

Sami,

I think we may be saying roughly the same thing.

The point I wanted to stress is that CLASSTYPE alone is useless if
you are using E-LSPs. I think this point is recognized by a number
of folks. As the WG drafts exist today, this object can only 
be used with L-LSPs. i.e. where you only need signal one set of
traffic parameters per LSP.


Nabil



> 
> Nabil:
> The problem you are describing is a generic problem
> with E-LSPs, not specific to CLASSTYPE (or DiffServ TE).
> When an E-LSP is signalled, the signalled bandwidth
> is associated collectively to the whole LSP, not
> individual PSCs. Therefore, it is not possible
> to do diffserv resource adjustments (e.g., queue bandwidth)
> based on the signalled bandwidth.
> The draft you are pointing out is a good first attempt
> to solve some of these issues. However, IMHO, E-LSPs and
> DiffServ TE concepts still require a lot of work.
> 
> Regards,
> --Sami
> 
> > -----Original Message-----
> > From: Nabil Seddigh [mailto:nseddigh@tropicnetworks.com]
> > Sent: Wednesday, November 07, 2001 11:29 AM
> > To: Iren, Sami
> > Cc: 'Vishal M'; mpls@UU.NET
> > Subject: Re: CLASSTYPE object
> >
> >
> > Sami,
> >
> > Currently, the CLASSTYPE object cannot really be used with
> > E-LSPs in the manner you describe. The CLASSTYPE object is only
> > useful with L-LSPs.
> >
> > In order to utilize the concept of bandwidth pools in E-LSP,
> > one needs to be able to signal traffic parameters per class
> > or per-PSC. That capability is not present in the standards today.
> >
> > The closest proposal to achieve what you describe is the following:
> > http://www.ietf.org/internet-drafts/draft-ganti-mpls-diffserv-
> elsp-00.txt
> 
> Best,
> Nabil Seddigh
> 
> >
> > Your understanding that the CLASSTYPE is used to
> > specify the bandwidth pool to be used for allocating
> > resources during LSP Setup is correct. In this context,
> > the service class which is signalled via EXP bits
> > (in the case of E-LSPs) dictate the treatment given
> > to the individual packets, and the CLASSTYPE, which
> > is signalled per LSP, dictates where the bandwidth
> > is coming from (used in CAC).
> >
> > [....deleted....]
> >
> > Ideally, I would prefer to get rid of the CLASSTYPE
> > concept and advertise available bandwidth per service class.