[Bloat] Rigorous Coffee Shop Bloat Testing

Dave Taht dave at taht.net
Tue Sep 3 23:17:51 EDT 2019


Kenneth Porter <shiva at sewingwitch.com> writes:

> On 9/3/2019 5:40 PM, Stephen Hemminger wrote:
>> There was a recent Wall Street Journal article that faster Internet doesn't mean anything.
>> https://www.wsj.com/graphics/faster-internet-not-worth-it/
>>
>> I just thought "faster Internet just exposes your existing Bufferbloat"
>
> I hit a paywall trying to read that so I looked up the article title
> and found some interesting commentary:
>
> https://tech.slashdot.org/story/19/08/20/1450204/the-truth-about-faster-internet-its-not-worth-it
>
> https://stopthecap.com/2019/08/20/wall-street-journal-says-faster-internet-not-worth-it-but-they-ignore-bottlenecks-and-data-caps/
>
> Most people are streamers and won't fill a fat pipe. The big winners

The market has shifted. With streaming becoming a thing (instead of
bittorrent), ISPs had to ensure that there was enough bandwidth during
peak hours across their entire platform to keep users from screaming.

Once everybody upgraded to "enough", because the usage pattern had
shifted, few want or need more.

> of fast Internet are people who want to download a huge game and play
> it quickly. But those are rare. (I'm a gamer but I'm patient and can

Well in the cases of shared internet, more BW (and less bufferbloat)
helps a lot. Small businesses such as coffee shops, etc. But they
are a much smaller portion of the market than the home.

It's semi-worse/semi-better than that - usage has shifted to people's
LTE phones for a lot of things, and they forget to enable wifi.

I think ISPs have shot themselves in the foot, perhaps permanently.

* By shipping buggy gear and bad wifi
* Not investing in good ipv4 and ipv6 technologies
* By having bad latency, the user experience is comparable to lte apps
* By not allowing services in the home, they've shifted to the cloud

Growth for ISPs could come from higher upload bandwidths - and I kind of
hope they start marketing "FASTER UPLOADS!" "Great Gaming!" "Killer
Videoconferencing!" "Multiple security cameras!" etc - Even a mere
doubling of upload speeds helps a lot.

... instead of doing that, some are trying to put on data caps and other
barriers to using the oversupply of bandwidth they now have with no
demand. Certainly there's a big focus on somehow delivering 4k or higher
video - and that's really not very perceptible, just a bunch a bits....

cisco just pulled out of the docsis 10Gbit effort. There's no demand.

The demand for (some) fiber exists simply because cablemodems are so
bad - and have 5x the baseline RTT as fiber does - and hype. Gfiber
and FIOS struggle. Sonic in SF, is doing well, but that's SF for you.

"DOCSIS-LL - now with lower RTTs! Buy it now!"

> wait a day to play so I'm happy to save money on a cheaper package
> that can be used for something else.)

I subsist on a really tiny amount of bandwidth, managed of course by cake.

>
> As you say, when people report slow Internet, it's probably bloat, not
> the speed of the package. But faster packages make money for the ISPs.

Nobody on the slashdot article chimed in on the bufferbloat front. But
whatever.

Back in 2012... I thought we'd basically hit "peak bandwidth" at
~40Mbit/user, and we just needed to optimize for RTTs to utilize that
always at multiple levels - be that physical rtt on link - or via
cdns - or wifi - etc.

I'd argued strenously with cablelabs that their benchmarks were wrong -
that web page growth was not exponential (avg web page size has only
doubled in 7 years) - this is a great resource:

https://httparchive.org/reports/state-of-the-web#bytesTotal

... that everything was bound by RTT - that a single queue AQM
was smoked by fq due to the reduced RTTs on bidirectional traffic...

google published a great benchmark about why RTT was so important...

https://www.igvita.com/2012/07/19/latency-the-new-web-performance-bottleneck/

and.... they didn't fix it. Took years to rollout docsis 3.1, which is
still only tiny.... better uploads rolled out first which helped on
generic RTTs....

Unless usage patterns change - fixing bufferbloat, and back to things
like torrent running all the time, or folk migrating services en-mass
back into their homes and businesses, I don't see bandwidth related
revenue growth for land-line ISPs. Sure, we'll see dsl users continue to
migrate to whatever they can, and people jump off of cable as soon as
they can get fiber... but that's due to the RTT more than the bandwidth.

I do hope - 10 years from now - someone points at what I just said and
laughs, showing their 10Gbit home fiber line saturated with brainwave to
smellovision full immersion bi-dir traffic, or something like that.

And I still hope people get out into the real world into more nice
coffeeshops, and hang out. I remember when lan parties were a thing....


More information about the Bloat mailing list