RE: IETF54- Informal discussion on BC Model for DS-TE

[Dimitry Haskin <dhaskin@axiowave.com>]

Francois,

Yes, if I had to choose between Russian Doll and Maximum Allocation models,
I would favor the latter. This is because I feel that potential QoS
degradation of a lower level service (probably the best-effort service
anyway) is preferable to arbitrary denying service to already established
flows (which by itself is an extreme case of service degradation of affected
flows).

Said that, I'm also thinking of the possibility of combining the concepts of
RDM and MAM in a single more flexible BC model with only slight complexity.
One such an approach would introduce an additional configuration parameter,
say 'nesting levels' (NL), that controls how many levels of the RD types of
bandwidth nesting is employed.

To illustrate how it may work, let's consider 3 CTs, say CT2, CT1, and CT0,
with BC2, BC1, and BC0 as respective bandwidth constraints.

Setting NL to 0 results in MAM, i.e.:
  - sum of CT0 reservations <= BC0
  - sum of CT1 reservations <= BC1
  - sum of CT2 reservations <= BC2

NL of 1 results in a combination of MAM and RDM:
  - sum of CT0, CT1, and CT2 reservations <= BC0
  - sum of CT1 reservations <= BC1
  - sum of CT2 reservations <= BC2

NL of 2 results in RDM:
  - sum of CT0, CT1, and CT2 reservations <= BC0
  - sum of CT1 and CT2 reservations <= BC1
  - sum of CT2 reservations <= BC2


This also depicted below:

NL=0:

|---BC2---->|---------BC1-------->|-------------------BC0----------->


NL=1:

|---BC2---->|---------BC1-------->
|-------------------BC0------------------->


NL=2:

|---BC2---->
|---------BC1-------->
|-------------------BC0------------------->


What do you think?

Regards,
  Dimitry


> -----Original Message-----
> From: Francois Le Faucheur [mailto:flefauch@cisco.com]
> Sent: Sunday, July 14, 2002 9:37 PM
> To: Dimitry Haskin
> Cc: te-wg@ops.ietf.org
> Subject: RE: IETF54- Informal discussion on BC Model for DS-TE
> 
> 
> Dimitry,
> 
> > >
> > > My impression is that it is just not possible to simultanesouly :
> > >          -(i) ensure bandwidth sharing (ie no bandwidth wastage)
> > >          -(ii) ensure bandwidth isolation (ie a CT cannot 
> have some of
> >its
> > > bandwidth taken by another CT)
> > >          -(iii) refuse to use preemption
> > >
> >
> >I respectfully disagree. I believe that (i) and (ii) can be achieved
> >simultaneously with (iii). Preemption of LSPs at a lower 
> service level
> >in order to allow LSP calls at a higher service level is not the only
> >way to achieve (i) and (ii). An alternative would be to allow a
> >potential service degradation of the already established 
> LSPs at a lower
> >service level without totally removing (often in an 
> arbitrary way) any
> >particular LSP from the service. Given my example, the sum 
> of normalized
> >CT1 and CT0 reservations would be allowed to exceed BC0.
> 
> Let's see if I understand exactly what you are proposing. If 
> I understand 
> correctly, you're arguing in favor of the Max Allocation 
> model where you 
> would configure the sum of the constraints much larger than 
> the total capacity.
> For example, on a link of 100, you would configure something like:
>          - Sum (CT0) <= BC0 = 100
>          - Sum (CT1) <= BC1 = 50
> And then you are saying:
>          - you can always make full use of the link (ie even 
> if there is no 
> CT1 you can setup 100 of CT0)
>          - you can always achieve 50 worth of CT1 without 
> preemption (even 
> if you have a 100 of CT0, you will accept 50 of CT1 and then 
> CT0 will just 
> get degraded).
> 
> Is this right? or are you proposing another BC model and if 
> so can you 
> detail it for us?
> 
> Thanks
> 
> Francois
>