Suggestions on draft-wlai-tewg-bcmodel

["Francois Le Faucheur (flefauch)" <flefauch@cisco.com>]

Waisum,

A couple of suggestions for your next rev:

1) Selection of BC values
==========================
Selecting BC values for the compared models (when the prime element of
comparison is LSP blocking) is clearly a key aspects of the comparison.
To illustrate that point:
	- If I configure MAM with BC1=BC2=BC3=1000 while I configure RDM
with BC1=30, BC2=20, BC3=10, it is pretty obvious that MAM will have
much lower blocking. 
	- Conversely if I configure MAM with BC1=BC2=BC3=30 while I
configure RDM with BC1=1000, BC2=900, BC3=800, it is pretty obvious that
RDM will have much lower blocking.
Now, the reason BC values cannot simply be arbitrarily inflated is
because they result in QoS degradation: if I configure  BCs=O(1000) on a
link of O(10), there will be QoS degradation across CTs.

Now, if I understand correctly, the approach you took for selecting BC
vales for RDM, was to pick BC values so that RDM would exhibit the same
blocking probabilities as MAM for a given reference LSP load. This gave
you :
	- for MAM: BC1=6, BC2=7 and BC3=15
	- for RDM: BC1=6, BC2=11, and BC3=15
And then you use this values to compare RDM and MAM blocking performance
when increasing the load beyond the reference LSP load. 

This does not seem to me like a good base for comparison, because these
values correspond to allowing much more QoS degradation for MAM than for
RDM. For example, these values would accept a total load of 6+7+15=28
with MAM , while the maximum load is limited to BC3=15 with RDM. 

An analogy would be where someone wants to compare a Porsche and the
cheapest-car-on-market in terms of performance going uphill. And to do
so, the first thing he would do is put a 3-ton-trailer beyond the
Porsche to make sure they have same performance on flat ground. Then ,
when comparing uphill performance, it may be that the
cheapest-car-on-market would perform better than the Porsche with its
3-ton-trailer. From this, we would conclude that cheapest-car-on-market
has better uphill performance than a Porsche. No?

So, I would more value in a comparison of MAM and RDM with BC values
which yields comparable QoS degradation.
Probably something like:
	- for MAM: BC1=6, BC2=7 and BC3=15
	- for RDM: BC1=6, BC2=13, and BC3=28
Could you use these sorts of BC values in your comparison?

Was there some rationale for the 3-ton-trailer that I missed?


2) MAM definition
==================
We have an open discussion on the list about the MAM definition, about
whether we should update the current definition to add an aggreagte
Bandwidth Constraint (on sum across all CTs) or not. This is something
we will need to discuss/conclude on, but in any case, I suggest you make
it explicit in your draft whether you apply an aggregate BC for MAM in
your analysis or not. (for example , currently you only mention 3 BC
values for 3 CTs so I assume you did not assume an aggregate BC). 


Thanks for considering this

Francois