Re: [Codel] fq_"codel" in apple's gear

[Dave Taht <dave.taht@gmail.com>]

On Sat, Oct 14, 2023 at 11:17 AM Jonathan Morton <chromatix99@gmail.com> wrote:
>
> > On 14 Oct, 2023, at 8:23 pm, Dave Taht <dave.taht@gmail.com> wrote:
> >
> > I would really like someone with a fresh set of eyeballs to review the
> > "codel"  portion of apple's latest code drop here:
> >
> > https://www.linkedin.com/feed/update/urn:li:activity:7118969694987935744/
>
> So far as I can tell, it still implements DRR++ (with a comment referencing RFC-8290), but not Codel (there's also no comment referencing RFC-8289).  There's at least one function for handling ECN which is simply not called, at least not in that source file.

Thank you for summoning the gumption to look again. I was having blood
pressure problems.

> In fact, I looked in the header to see where an ECN mark counter was defined.  It is referenced in essentially three places:  the definition, an extraction for reporting to userspace, and an increment - in a function implementing L4S marking, not Codel.
>
> A similar counter for "early dropped" packets is incremented only when packets have to be dropped in an emergency because the queue is *full*.  That's the other counter I would expect to be touched by an AQM implementation.

Yep. I had run some tests then and saw the rtts climbing alarmingly,
while polling their stack and only seeing drops in big bursty batches,
total resets of the tcp flow, and so on.

> So, no tangible improvement that I can find easily.
>
>  - Jonathan Morton



-- 
Oct 30: https://netdevconf.info/0x17/news/the-maestro-and-the-music-bof.html
Dave Täht CSO, LibreQos