[Bloat] ONTs and ITU - T G.988

[Dave Taht]

What appeared to be the case was that a ONT had a 50ms buffer at
100Mbit and was reconfigured to drive a gig and thus only had 5ms.

At least, that's what the data looked like. This was on a fairly
recent reddit thread a month or two back, that I cannot find right
now. I'm pretty sure it was in the context of eero's
rolling out a working htb+fq_codel implementation finally for the eero
6, (or maybe it was the comcast AQM?0 and us discovering it didn't do
any good, that the ont was seemingly brutally underbuffered, which
took days of scratching my head.... multiple flent tests... wish I
could find that thread...

This is kind of ironic, but the structure of the speedtest (8+ flows
with short RTTs) actually could lead naturally to buffers that short
and few would notice.

Anyway, that sparked me finally going and reading up a bit more about
how ONTs work...


On Wed, Jan 12, 2022 at 4:40 PM Marco Belmonte
<marco at heavenlysanctuary.com> wrote:
>
> Dave,
>
> Can you describe to me what under buffered would look like? Let's say
> I'm playing Rocket League on a Frontier FIOS connection... When a
> situation arises where the connection is under buffered you might see
> __<Insert DT's Big Brain words here>___.
>
> A lot of interrupts being generated? connections timing out? lag?
>
> I'm guessing that many of the same symptoms that accompanying buffer
> bloat also manifest themselves with buffer bleed?
>
> Marco
>
>
> On 1/12/2022 3:42 PM, Dave Taht wrote:
> > Up until this point I haven't cared all that much about fiber, as most
> > of my observations were that it had "reasonable" buffering and all I
> > could hold in my head was the 802.11 standards. Since cablemodems are
> > well on their way to being fixed, wifi is looking good, and I recently
> > came across an ONT that was actually underbuffered, my holiday reading
> > consisted of pounding through a ton of fiber specs like this one:
> >
> > https://www.itu.int/rec/dologin_pub.asp?lang=e&id=T-REC-G.988-201711-I!!PDF-E&type=items
> >
> > There's a lot of promising info in this spec starting from page 104.
> >
> > My 1st question is - are there any ONTs that actually do do pause
> > frames? Or providers that configure for them?
> >
> > My second is - what is a cheap way to setup a lab to emulate a good,
> > common, version of gpon e2e?
> >
> > Third - what OS do these things run? Who makes a "good" one?
> >
> _______________________________________________
> Bloat mailing list
> Bloat at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/bloat



-- 
I tried to build a better future, a few times:
https://wayforward.archive.org/?site=https%3A%2F%2Fwww.icei.org

Dave Täht CEO, TekLibre, LLC