Re: [tcpm] CUBIC rfc8312bis / WGLC Issue 3

[Markku Kojo <[e-mail]>]

Hi Neal,

thank for your proposal. Note that I moved your proposal here under the 
thread on Issue 3 which it is related to (see below).

On Tue, 14 Jun 2022, Markku Kojo wrote:

> Hi all,
>
> this thread starts the discussion on the issue 3: the incorrect way of
> setting alpha to 1 when in the fast convergence mode within the Reno-friendly 
> region (Issue 3 a) and the issue of setting Wmax for a congestion event 
> arriving when in slow start (Issue 3 b):
>
>
> 3 a) The rule for changing alpha to 1 when Wmax is reached in the
>     Reno-friendly region is the correct thing to do during the normal
>     steady state. However, it is incorrect action to take when in the
>     fast convergence mode within the Reno-friendly region because it
>     would act just *opposite* to what CUBIC should do when in the fast
>     convergence mode; instead of slowing down the increase rate during
>     congestion avoidance it actually accelerates because alpha becomes
>     increased to 1 earlier than when not in the fast convergence mode.
>     This seems an obvious mistake with the quite recent modifications
>     to the rfc8312bis.

On Mon, 6 Jun 2022, Neal Cardwell wrote:

> Markku, thank you for your clear and comprehensive write-up.
>
> Regarding this particular, somewhat easier point:
>
[snip]
> I agree this seems like a clear and straightforward thing to fix.
> Indeed, my current draft Linux TCP CUBIC patch series for this
> "changing alpha to 1" mechanism already uses your suggested approach,
> and it seems to behave sensibly.
>
> I have posted a proposed pull request to try to encode this in the
> draft:
>   https://github.com/NTAP/rfc8312bis/pull/146
>
> Please take a look and see what you think. Thanks!
>
> neal

This seems to correct the issue, thanks. It would be useful to introduce 
prior_cwnd in section 4.1.2 as already discussed in the github pull 
request. Also needs text specifying the initialization in other 
case than "congestion event". This also seemed to attention in 
github.

Note that this fixes the problem of CUBIC taking an action opposite what 
it should take. Even after fixing this, there is no "fast convergence" 
for CUBIC when it is in the Reno-friendly region, as the fast convergence 
only has some effect when CUBIC operates in the "genuine" CUBIC mode.
But this issue relates to Issue 4 and discussion on it should be continued 
there.

Thanks,

/Markku

> 3  b) Wmax needs to be set differently for a congestion event arriving
>     when in slow start and when in congestion avoidance
>     If Wmax is set for a congestion event arriving when in slow start
>     like currently specified, it results in Wmax with a value that is
>     two times higher than the intended Wmax value (the co-authors who
>     are the original developpers of CUBIC have agreed on this).
>
> Thanks,
>
> /Markku
>