[Bloat] [LibreQoS] [Starlink] Researchers Seeking Probe Volunteers in USA

Dave Taht dave.taht at gmail.com
Mon Jan 9 13:33:05 EST 2023

On Mon, Jan 9, 2023 at 9:05 AM Jeremy Austin <jeremy at aterlo.com> wrote:
> On Mon, Jan 9, 2023 at 10:26 AM Dave Taht via LibreQoS <libreqos at lists.bufferbloat.net> wrote:
>> I
>> 2) Few are actually testing within-stream latency
> If some kind of consensus can be generated around how latency under load should be reported, and bearing in mind that to date Preseem measures non-destructively, i.e., not generating synthetic flows, we would be happy to help by adding that analysis to our regular reporting.

Yes, it is presently too vague a term. What load? (and my kvetch,
mostly - over what time period)?

> We have some FWA-specific latency numbers in our reports, but will be adding more granular reporting for other access tech as well. A single-dimension histogram isn't sufficient, IMO, but do we really need to teach everyone to read CFS? Maybe.

In writing a really ranty blog entry about my new chromebook over the
holiday (feel free to subject yourself here:
https://blog.cerowrt.org/post/carping_on_a_chromebook/ ) I realized
how different my workloads were than most, and why latency under load
matters so much to me(!) -

I regularly use ssh from the front of my boat to aft, suffer from
running out of LTE bandwidth, use X to remotely screen share, do big
backups, git pulls and pushes, live 24/7 in 15+ mosh terminal tabs to
machines all over the world, play interactive network games, and do
massive compiles of huge source code bases.

I realized, today, after venting my spleen in that blog, that it was
highly unlikely that the vast majority of people out there used their
networks as I do, and it was irrational of me to project my needs on
theirs. Despite identifying new applications, like cloud gaming, and
edge computing, that would benefit if we smashed the LUL there, I am
selfishly in this game to make my DISPLAY variable "just work" for
emacs as well as it did in the 90s.

But then again, I'm pretty sure, most, at least occasionally, push a
big file up or down, at the very least, and get burned by bufferbloat.
The subset of gamers and to some extent videoconferencers, also - but
the majority?

So a really interesting piece of data that I'd like to acquire from an
ISP-facing network is not even the bloat, but, a histogram of the
durations from syn to fin for more normal users.  I imagine that 99%
of all tcp transactions to/from home users to be very, very short.
Uploads, longer.

> --
> Jeremy Austin
> Sr. Product Manager
> Preseem | Aterlo Networks
> preseem.com
> Book a Call: https://app.hubspot.com/meetings/jeremy548
> Phone: 1-833-733-7336 x718
> Email: jeremy at preseem.com
> Stay Connected with Newsletters & More: https://preseem.com/stay-connected/

This song goes out to all the folk that thought Stadia would work:
Dave Täht CEO, TekLibre, LLC

More information about the Bloat mailing list