Re: [Tsvwg] RE: HEADS UP: consensus call on draft-floyd-tsvwg-cc-alt

[Sally Floyd <[e-mail]>]

Mitchell -

My apologies for the delay in replying.

> 	My two cents..
>
> 	One cent
> 	---------
> 	Is TCP fair? With two flows that start at the same time, with
> 	at least 1 hop in common, doesn't a lower RTT flow, absorb
> 	a greater amount of bandwidth? Of course it does..
>
> 	Shouldn't TCP allow both, in this example, to achieve a
> 	stasis window where both flows are using approximately
> 	50% of the available bandwidth on this common hop?
>
> 	There are more examples of where TCP isn't fair. However,
> 	should TCP have a priority goal of fairness? If one flow
> 	sees congestion before another and does a MD, where the
> 	other flow keeps doing AI, then this does not violate the
> 	TCP protocol.
>
> 	Maybe I missed it, but should the doc discuss the effects
> 	of tail-drop versus RED on multiple flows?
>
> 	Thus, is it fair to measure different TCP implementations
> 	in a heterogenous environment based on some set of criteria
> 	of fairness?? Shouldn't we EXPECT some level of UNFAIRNESS
> 	in heterogeneous environments (multiple different systems
> 	using diifferent TCP implementations)?
>
> 	Two cent
> 	--------
> 	Maybe I missed it, but should the doc discuss the effects
> 	of tail-drop versus RED on TCP flows? Which one is
> 	"more fair"? Does RED reduce congestion as it reduces
> 	the effective size of the input buffers, which allows
> 	for bursty traffic. Isn't the TCP protocol bursty in
> 	nature due to the fact that no interpacket gap is
> 	MANDATORY or even suggested  between segments on
> 	transmission?

As I said in my earlier email:

>> draft-floyd-tsvwg-cc-alt doesn't attempt to give detailed 
>> *requirements*
>> for fairness, it just discusses fairness (flow-rate fairness) as one 
>> of the
>> issues to be considered in evaluating new congestion control 
>> mechanisms.
>>
>> Detailed metrics on fairness, issues about fairness and packet size, 
>> etc.,
>> are discussed in draft-irtf-tmrg-metrics-07, Metrics for the 
>> Evaluation of
>> Congestion Control Mechanisms.  That draft, and an updated version
>> in progress that has not yet been submitted, are available from
>> "http://www.icir.org/floyd/papers.html".

That seems to me the right document for a discussion of all of the 
intricacies of
fairness as a metric.  That document discusses issues about fairness 
and RTT,
and a number of other issues.

- Sally
http://www.icir.org/floyd/