Re: [iccrg] LEDBAT++, rLEDBAT, and slowdowns

I think those are two different kind of unfairness effects. One can lead to
starvation, the other to biased fair shares in favor of small RTTs.

On Tue 10 Sep 2019 at 11:46, Michael Welzl <michawe@ifi.uio.no> wrote:

>
>
> > On 10 Sep 2019, at 10:22, Luca Muscariello <muscariello@ieee.org> wrote:
> >
> > I think, Roland is right about fairness.
> > More generally AIAD is unfair, which turns out to be a stability issue.
>
> That said, I hope people are aware that AIMD isn't really much better in
> that respect. This gets quite clear when simply looking at some
> trajectories with unequal RTTs.
> Easy and intuitive to do with this tool:
> http://heim.ifi.uio.no/~michawe/research/tools/cavt/index.html
>
> Cheers,
> Michael
>
>
> > This is also known for Vegas.
> >
> > So even if RTT_min is known AIAD is a bad design choice.
> > Still it remains to infer RTT_min.
> >
> > There is a reference that I report below that can be useful to fix
> LEDBAT for what concerns
> > intra/inter protocol fairness seen from a stability perspective.
> >
> > A. Tang, X. Wei, S. H. Low and M. Chiang,
> > "Equilibrium of Heterogeneous Congestion Control: Optimality and
> Stability,"
> > in IEEE/ACM Transactions on Networking, vol. 18, no. 3, pp. 844-857,
> June 2010.
> > doi: 10.1109/TNET.2009.2034963
> >
> >
> >
> > On Tue, Sep 10, 2019 at 9:36 AM Bless, Roland (TM) <roland.bless@kit.edu>
> wrote:
> > Hi Marcelo,
> >
> > Am 09.09.19 um 17:34 schrieb marcelo bagnulo braun:
> > > I am not sure i understand your answer though. I mean the whole problem
> > > of LEDBAT latecomer advantage is that the late comer misestimates the
> > > base delay (i.e. RTT_min?) and then it adds the target delay on top of
> > > the wrongly estiamted base delay.
> >
> > Yes.
> >
> > > So, i dont understand what do you mean that the mechanism assumes
> > > knownledge of the base delay. Clearly, i am missing something here,
> > > could you educate me?
> >
> > As Luca wrote, their proposed algorithm requires a _perfect knowledge_
> > of RTT_min, i.e., transmission and propagation delay
> > without any queueing delay.
> >
> > Note that, even if one would have perfect knowledge of queueing delay,
> > the latecomer advantage would turn into a latecomer disadvantage,
> > because the latecomer would back-off. Fairness remains still
> > an issue then...
> >
> > Regards
> >  Roland
> >
> > _______________________________________________
> > iccrg mailing list
> > iccrg@irtf.org
> > https://www.irtf.org/mailman/listinfo/iccrg
> > _______________________________________________
> > iccrg mailing list
> > iccrg@irtf.org
> > https://www.irtf.org/mailman/listinfo/iccrg
>
>