Re: adjusted_rtt

[Ian Swett <[e-mail]>]

Good catch Timo, I think you're correct.

PR sent: https://github.com/quicwg/base-drafts/pull/4861

On Thu, Apr 8, 2021 at 4:15 AM Lars Eggert <[e-mail]> wrote:

> I created https://github.com/quicwg/base-drafts/issues/4860 to track this.
>
> Thanks,
> Lars
>
>
> > On 2021-4-7, at 20:22, Timo Völker <[e-mail]> wrote:
> >
> > Hi,
> >
> > I noticed in a simulation with a constant RTT that the smoothed_rtt
> value includes the ack_delay. It looks like my implementation follows the
> recovery draft. By looking closer, I believe I found a tiny error in the
> draft.
> >
> > Section 5.3 contains this sentence.
> >
> > "MUST NOT subtract the acknowledgment delay from the RTT sample if the
> resulting value is smaller than the min_rtt."
> >
> > I interpret this as do not subtract ack_delay from latest_rtt if
> >
> >  latest_rtt - ack_delay < min_rtt
> >
> > Or to say the inverse, substract ack_delay from latest_rtt if
> >
> >  latest_rtt - ack_delay >= min_rtt
> >  <==> latest_rtt >= min_rtt + ack_delay
> >  <==> min_rtt + ack_delay <= latest_rtt
> >
> > The pseudo code in Section 5.3 and in the appendix use a smaller instead
> of an equal or smaller in the if statement.
> >
> >  if (min_rtt + ack_delay < latest_rtt):
> >
> > It should be <= or did I miss something?
> >
> > Timo
>
>