Re: [rmcat] test cases for coupled cc [was: Re: Review of draft-ietf-rmcat-eval-test-02]

[Zaheduzzaman Sarker <[e-mail]>]

On 2016-02-25 17:20, Mirja Kühlewind wrote:

>>>
>>> I agree with Michael. My view is here that we need to evaluate each
>>> combination of a certain scheme to couple things using a certain
>>> congestion control algorithm separately.
>> Fine. but again what exactly are we supposed to evaluate? the
>> performance of the congestion control algorithm or the performance of
>> the coupled congestion control algorithm? or something else.
>
> The performance of the combined system of the coupling algo and the
> selected cc aglo.
ok, the current draft says "This document describes the test cases
    to be used in the performance evaluation of those candidate
    algorithms."
as now we are talking about combined system, this needs to be changed I 
guess. right? but we need to clarify what this combined system means, I 
hope we are not leaning towards the whole RTC system evaluation.
>
>>> That means, if someone proposes a new (generic) scheme for coupling,
>>> we as a group would only advise to use this scheme with a certain cc
>>> algorithm if we have seen results for this specific combination. Does
>>> this make sense to you?
>> Kind of, this means it is fine to have a coupled congestion control that
>> may not work will all the combinations even if RMCAT proposes more that
>> one congestion control algorithms.
>
> Yes (as long as this is clearly stated in the doc).
ok. but the eval-test draft doesn't need to clarify that, right?
>
>>
>> I didnt get the "generic"  part though. How can the WG call a coupled
>> congestion "generic" even if that fails to show results with all the
>> RMCAT CC algorithms (assuming RMCAT has proposed more than one CC
>> algorithms)?
>
> That's only a matter of how it is described in the doc. Basically you
> say, we describe this in a generic way, but it's only tested together
> with these algos. If you want use it with others, it might be possible,
> but you have to (adapt and) test it on your own. Important is that you
> have both in the same doc: the generic description and how it is applied
> to certain algos. Does that make sense? Or it this a problem for any of
> your work?
Ok so here "generic" does not mean the algorithm is generic, it points 
to the generic description of the coupled CC.

No problem for any work as long as the above clarification text is not 
part of the eval-test draft.

>
> I believe that is at least the approach we are currently taking for the
> coupled-cc draft that we have. Micheal, please speak up if you have a
> different view here!
>

BR
-- 
Zahed

==================================================
ANM ZAHEDUZZAMAN SARKER


Ericsson AB
Services, Media and Network Features
Laboratoriegränd 11
97128 Luleå, Sweden
Phone [phone]
Fax [phone]
SMS/MMS [phone]
[e-mail]
www.ericsson.com

==================================================