[tcpm] Comments on CUBIC slides for tcpm meeting on Fri

[Markku Kojo <[e-mail]>]

Hi all,

I browsed quickly the slide set for rfc8312bis.

A few comments about the issues as the text on slides does not 
quite capture all issues that well.


Slide 2, Issue 2:

It is not that doubling in slow start makes 0.5 convenient, but it makes 
it necessary.

Beta = 0.5 is designed such that the buffer *is full* after the decrease 
in slow start. Therefore, using Beta > 0.5 results overload, i.e, 
unnecessarily injecting packets that are only dropped by a tail-drop 
router due to overload.

Please see the reasoning for Beta = 0.5 in [Jacob88].

We have tons of studies and over three decades of experience for Beta= 0.5 
in slow start, so for a standards track RFC we don't need any additional 
testing. It is already the currently specified draft standard (RFC 5681). 
Instead, deviating from Beta=0.5 when in slow start is in conflict with 
RFC 5681 and requires justification. The draft does not give any.
Nor does it say what potential benefits it would bring. AFAICT nothing, it 
only hurts the flow itself. Of course, being too aggressive during 
recovery from slow-start overshoot when competing with other flows will 
potentially give unustfied gain for the flow as it steals the capacity 
from others.

[Jacob88] V. Jacobson, Congestion avoidance and control, SIGCOMM '88.

Issue 4 and 5 are missing. I've recently sent a separate description for 
the issue 4. For issue 5, we have discussion in github and summary sent 
earlier. I try to send a separate issue description to the wg list 
shortly I am not able to send it before the wg meeting.

Thanks,

/Markku