Concern with <draft-wlai-tewg-bcmodel-02.txt>

Waisum and all,

I would like to share with the group the concern I have expressed to
Waisum on current version of <draft-wlai-tewg-bcmodel-02.txt>.

Clearly, when looking at BC Model performance purely from the viewpoint
of LSP blocking/preemption, the choice of the BC values is key. If I
arbitrarily increase the BC values, I can accept more LSP and reduce (or
completely eliminate) blocking and preemption. The reason those can not
simply be arbitrarily increased is because they would result in
potential QoS degradation (ie accepting more LSP traffic than the
correponding Diff-Serv scheduler can service).

So, it seems most relevant to compare models using BC values which
correspond to similar QoS degradation (on the higher priority classes)
in both models. For example, MAM(6/7/15) should actually be compared
with RDM(6/13/15). This is because [assuming fairly typical Diff-Serv
scheduler configuration so that if bandwidth is not used by EF queue ,
most of it can be allocated to "Premium Data" queue] you can indeed, in
case there is no EF LSP, accept up to 13 worth of Premium LSP, without
any QoS degradation on Premium traffic. (Really, this is a KEY point of
RDM: be able to have the unused bandwidth of class N being reallocated
in priority to Class N+1).
So, the bottom line is that both MAM(6/7/15) and RDM(6/13/15) behave
very much the same (ie equally well) in terms of QoS degradation for
Voice+Premium.

However, draft-wlai-tewg-bcmodel-02.txt currently compares MAM(6/7/15)
with RDM(6/**11**/15) and the justification for arbitrarily picking
these values of BCs is so that the two models happen to have equal
blocking/preemption under the reference load. Then this same BC values
are used when increasing the load. I don't see the rationale for saying:
"in order to compare two models let us start by tweaking one down so
they both have the same performance level". As I explained earlier on
this list, to me this is like deciding to attach a 3-ton trailer behind
a Porsche so it has the same performance on flat ground as a low-end
vehicle, and then conclude that the low-end vehicle has higher
performance than a Porsche going uphill because it behaves better than
the Porsche with its 3-ton trailer.

Another way to look at this is the following:
I think we agree that BC models performance is a multi-dimension thing
(one dimension being blocking/preemption, another dimension being QoS
degradation of premium classes, ...). While there is some merit in
comparing performance of BC models in one single dimension (ie
blocking/preemption) and ignoring other dimensions (ie QoS degradation),
this is only meaningful if the parameter settings are such that the
various models are on equal footage in the ignored dimensions (eg QoS
degradation).

My objective is not to tweak comparison so that RDM systematically has
better performance than MAM in all cases. We know that MAM behaves
better for example in terms of class isolation when preemption is not
used. But I think the comparison needs to put models on equal footage
and should reflect one of the key benefit of RDM (again this
reallocation of bandwidth from Class N to Class N+1).

Would it be possible for <draft-wlai-tewg-bcmodel-02.txt> to use BC
values 6/13/15 for RDM (instead of 6/11/15) when comparing to
MAM(6/7/15)?

Thanks

Francois