Re: [tcpm] [tsvwg] draft-ietf-tcpm-rfc8312bis-04: flipped exponents in the AVG_W_cubic formulas?

[Bob Briscoe <[e-mail]>]

Lars, (and switching again to tcpm from tsvwg)

On 03/09/2021 19:53, Lars Eggert wrote:
> This was my mistake when doing the formula - THANK YOU for finding that!

[BB] A nit (I wouldn't have bothered saying this if the draft hadn't 
just become a candidate for a major drains-up following Markku's review).

FWIW, I prefer the style of formulae as they were in RFC8312.
I'm not going to insist, but trying to lay out formulae like ASCII art 
didn't work for me. Sorry, I know it must have taken some time to lay it 
all out, which might not have been in vain if I'm overruled. Also, I 
don't think I've seen any other RFCs attempt to do this.

Showing formulae with superscripts and subscripts and fractions on 
multiple lines sometimes made it harder not easier to visualize the 
formulae (for me).
But the main disadvantage is that it's not easily machine-readable. So 
it cannot be accurately translated into different formats and it's more 
ambiguous. It also creates problems when quoted in emails if it gets 
quoted in a variable width font (which just happened to me, hence this 
email).

Admittedly, using ^ and _ for superscripts and subscripts is not 
brilliant either. Nor is using () / () for fractions. Nor sqrt() etc. 
But at least it's what people are used to in code, and in mark-up 
languages and LaTeX.

See also RFC1003, RFC1019, draft-ash-alt-formats, etc.

Bob

PS. I'm guilty of hypocrisy in that an RFC I co-authored shows complex 
fractions as ASCII art, altho the numerator and denominator are in 
C-code style maths. See 
https://datatracker.ietf.org/doc/html/rfc7893#page-9
FWIW, I didn't write the parts with the formulae in.

>
> -- 
> Sent from a mobile device; please excuse typos.
>
>> On Sep 3, 2021, at 21:39, Lisong Xu <[e-mail]> wrote:
>>
>> 
>> Thank you very much, Neal, for finding this important typo. You are 
>> right that the exponent should be 3/4 instead of 4/3 for Figures 5, 
>> 6, and 7.
>>
>> Thanks
>> Lisong
>>
>> ------------------------------------------------------------------------
>> *From:* Neal Cardwell <[e-mail]>
>> *Sent:* Friday, September 3, 2021 1:01 PM
>> *To:* [e-mail] 
>> <[e-mail]>
>> *Cc:* tsvwg IETF list <[e-mail]>
>> *Subject:* draft-ietf-tcpm-rfc8312bis-04: flipped exponents in the 
>> AVG_W_cubic formulas?
>> Non-NU Email
>> ------------------------------------------------------------------------
>> Dear rfc8312bis authors,
>>
>> First, thanks for all your work on rfc8312bis!
>>
>> Next: I was about to use the AVG_W_cubic formulas in the rfc8312bis 
>> text, and noticed what appears to be a discrepancy between the 
>> rfc8312bis AVG_W_cubic formulas and the AVG_W_cubic formulas in both:
>>
>> (1) the RFC 8312: *MailScanner has detected a possible fraud attempt 
>> from "urldefense.com" claiming to be* 
>> https://datatracker.ietf.org/doc/html/rfc8312#section-5 
>> <https://urldefense.com/v3/__https://datatracker.ietf.org/doc/html/rfc8312*section-5__;Iw!!PvXuogZ4sRB2p-tU!XrfKDCjwS6huhjdCHYDYUlgBpZ4GD0kXzOG363gPaoXUS9UNHjoMt0_BK0Ny1Q$>
>> (2) the CUBIC paper: *MailScanner has detected a possible fraud 
>> attempt from "urldefense.com" claiming to be* 
>> https://dl.acm.org/doi/10.1145/1400097.1400105 
>> <https://urldefense.com/v3/__https://dl.acm.org/doi/10.1145/1400097.1400105__;!!PvXuogZ4sRB2p-tU!XrfKDCjwS6huhjdCHYDYUlgBpZ4GD0kXzOG363gPaoXUS9UNHjoMt0_wZ47m5g$>
>>
>> In both (1) and (2) the formulas have RTT and p taken to the power of 
>> 3/4; for example:
>>
>>     With beta_cubic set to 0.7, the above formula is reduced to:
>>         AVG_W_cubic = (C*3.7/1.2)^0.25 * (RTT^0.75) / (p^0.75) (Eq. 6)
>>
>> By contrast, the rfc8312bis text, in Figures 5, 6, and 7, has RTT and 
>> p taken to the power of 4/3:
>>
>>                                    ________________      ____
>>                                   /C * (3 + β     )   3 /   4
>>                              4   /           cubic    |/ RTT
>>                 AVG_W      = |  /  ---------------- * -------
>>                      cubic   | /   4 * (1 - β     )       __
>>                              |/              cubic     3 / 4
>>                                                        |/ p
>>
>>                                    Figure 5
>> ----
>>                                                    ____
>>                                       _______   3 /   4
>>                                   4  /C * 3.7   |/ RTT
>>                      AVG_W      = | / ------- * -------
>>                           cubic   |/    1.2         __
>>                                                  3 / 4
>>                                                  |/ p
>>
>>                                    Figure 6
>> ----
>> 3 / 4
>>                                              |/ RTT
>>                         AVG_W      = 1.054 * -------
>>                              cubic               __
>>                                               3 / 4
>>                                               |/ p
>>
>>                                    Figure 7
>> That is, it appears there are typos in the formulas in rfc8312bis 
>> that systematically swapped the 3 and 4 in the exponents for RTT and 
>> p, yielding exponents of 4/3 instead of the  3/4 value from the 
>> original paper and RFC 8312.
>>
>> best regards,
>> neal
>>

-- 
________________________________________________________________
Bob Briscoe                               http://bobbriscoe.net/