[CCWG] Re: ABE and CUBIC parameters beyond beta

[Kazuho Oku <[e-mail]>]

2026年8月19日(水) 20:21 Michael Welzl <[e-mail]>:

>
>
> On 18 Aug 2026, at 20:18, Yoshifumi Nishida <[e-mail]> wrote:
>
> Hi Michael,
>
> On Tue, Aug 18, 2026 at 5:46 AM Michael Welzl <[e-mail]> wrote:
>
>>
>>
>>> In any case, Yes. If we want to make Reno with decrease factor 0.8
>>> (which is AIMD(1, 0.8)) and AIMD (alpha, 0.85) be the same, alpha will be
>>> 0.729.
>>>
>>
>> Yoshi, Kazuho: I must admit I’m lost with the maths here. What did you do
>> with which equations in RFC 9438 to arrive at this?
>>
>
> This actually came back from time to time during the discussions in the
> past.
> That's one motivation for me to write
> https://datatracker.ietf.org/doc/draft-nishida-ccwg-standard-cc-analysis/
>  I put a simple explanation for this in Appendix of the doc.
> I am thinking it might be good to publish this part of the explanations
> (could be an independent stream?) as a tip for discussions.
> --
> Yoshi
>
>
> Many thanks!  I get it now. Yes I agree that this would be useful to
> publish!   but why only this part - have you dropped the ball on
> the draft-nishida-ccwg-standard-cc-analysis draft, seeing that it hasn’t
> been updated since 2024?  That’s a shame, I think the whole draft is a
> useful document, in general.
>
> (sorry, perhaps the group’s discussion went negative?  I missed it, or
> forgot)
>
>
> Now that I understand Kazuho’s statement:
>
> > In the case where Reno uses 0.8 and CUBIC using 0.85, it should be (1 +
> 0.8)(1 - 0.85) / (1 - 0.8)(1 + 0.85) = 0.729.
>
> I want to add:
> 1) it misses a bracket, this should be  (1 + 0.8)(1 - 0.85) / ( (1 -
> 0.8)(1 + 0.85) )
> 2) since the result is 0.7297, I think it should be rounded to 0.73.
>

Agreed to both.

To explain the formula differently, Section 4.3 of RFC 9438 defines αcubic
as:

    αcubic = 3 * (1 - βcubic) / (1 + βcubic)

but the constant (3) is actually a precomputation of:

    (1 + βreno) / (1 - βreno)

with βreno set to 0.5. Compared to the formula above, the numerator and the
denominator are reversed because we're calculating the relative ratio
between reno and cubic.

With ABE, we use different βs, and therefore the ratio between αs needs to
be adjusted accordingly.


>
> Cheers,
> Michael
>
>

-- 
Kazuho Oku