RE: WG last call: draft-ietf-tewg-diff-te-reqts-04.txt

["Sudhakar Ganti" <sganti@tropicnetworks.com>]

Francois,

For the DS-TE, if the LSPs are setup based directly
on bandwidths BC0 to 7, then I understand about
the problems one may face. In this case, we don't know 
how much bandwidth is available for each class as we 
don't know the BC model. But we are using TE-Class 
bandwidths which would be derived from BCs and advertised. 
So irrespective of the BC model, if TE-Class available 
bandwidth is computed correctly, I do not see the need to
advertise what BC model each LSR uses. 

Can you also explain how an LER setting up the paths
may use the knowledge of BC model? Besides, if an LSR
supports multiple BC models, how is the predictability/ 
usefulness solved for the network operator unless we
enforce the same model across the domain? 

On the otherhand, enforcing the same BC model in a domain
may be very limiting to network operators. For example,
one may want to use bandwidth partitioning on some heavily
used links (to reduce call blocking for certain traffic 
aggregates for example) while use bandwidth sharing on 
less utilized routes. But by enforcing the same model, 
we are taking away this flexibility. Are network operators 
OK with this?

Regards
-Sudhakar

> -----Original Message-----
> From: Francois Le Faucheur [mailto:flefauch@cisco.com]
> Sent: Tuesday, June 04, 2002 12:30 PM
> To: Sudhakar Ganti
> Cc: te-wg@ops.ietf.org
> Subject: RE: WG last call: draft-ietf-tewg-diff-te-reqts-04.txt
> 
> 
> Ganti,
> 
> At 11:44 04/06/2002 -0400, Sudhakar Ganti wrote:
> >Hello,
> >
> >It is not clear to me what the specification of
> >bandwidth constraint (BC) model brings to DS-TE solution
> >as described in this draft. Current solution assumes
> >that each LSP belongs to a TE-Class and the constraint
> >based routing uses available bandwidth for TE-Class
> >that is advertised in computing a path. That is all
> >what is needed. BC information has local significance
> >in computing the TE-Class available bandwidth, but need
> >not be propagated to other nodes.
> >
> >Depending upon the network design and traffic matrix, different
> >bandwidth sharing models may be deployed in a network. Do we want to 
> >enforce the same model on every link across the whole domain
> ? What if
> >the models are different? Do we want to reject the path setup? 
> >Besides, the LSP setup is based on the advertised available bandwidth

> >of TE-Class. Is it not sufficent that whatever is the bandwidth 
> >model, as long as one computes correct available bandwidth for 
> >TE-Class and advertises, we can setup paths correctly? How would one 
> >use the knowledge of BC model for this purpose when constraint based 
> >routing is purely based on available bandwidth per TE-Class?
> >
> >Different BC models may have different implications of
> >call blocking etc in various scenarios. But that should
> >be left to network design and we should not be specifying what a 
> >default model should be. Is there any
> interoperability issue in
> >the network if different models are used?
> 
> If one LSR believes that BC0 is a contraint for CT0 while another LSR
> believes BC0 is a constraint for CT0+CT1, it will be very hard for a 
> network operator to achieve anything useful/predicatble.
> 
> Right now we are not sure if one or multiple BC models will be useful.
> But it seems clear that to achieve deployable interoperable
> DS-TE, all LSRs 
> must have at least one BC model in common. This is the one we 
> refer to as 
> the default model. Then all implementations will be free to 
> support many 
> other models, but at least all DS-TE implementation will support the 
> default BC model so they can work together.
> 
> Does it clarify?
> 
> Cheers
> 
> Francois
> 
> 
> 
> >Actually
> >I don't even see the need to advertise this as being specified as
> >optional IGP TLV in the DS-TE protocol draft.
> >
> >Therefore, a discussion on BC models in the draft
> >(e.g., various models and their performance) is fine. But
> >I don't see a need for "mandating" or specifying what
> >the default model should be and the corresponding
> unnecessary protocol
> >extensions.
> >
> >Any comments??
> >
> >Regards
> >-Sudhakar
> >
> > > -----Original Message-----
> > > From: Jim Boyle [mailto:jboyle@pdnets.com]
> > > Sent: Sunday, May 19, 2002 10:04 PM
> > > To: te-wg@ops.ietf.org
> > > Subject: WG last call: draft-ietf-tewg-diff-te-reqts-04.txt
> > >
> > >
> > >
> > > This message begins a WG last call for the following:
> > >
> > >       draft-ietf-tewg-diff-te-reqts-04.txt
> > >
> > > This WG document addresses a WG milestone.  It is proposed as an
> > > Informational RFC.  WG last call ends May 31, 2002.
> > >
> > >
> > >
> > >
> > >
> 
>