[Bloat] [Cerowrt-devel] closing up my make-wifi-fast lab

Dave Taht dave.taht at gmail.com
Sat Aug 25 17:22:09 EDT 2018


On Sat, Aug 25, 2018 at 1:04 PM David P. Reed <dpreed at deepplum.com> wrote:
>
> WiFi is a bit harder than IP. But you know that.
>
> I truly believe that we need to fix the phy/waveform/modulation space to really scale up open wireless networking capability. LBT is the basic bug in WiFi, and it is at that layer, melow the MAC.
>
> I have tried for 20 years now to find a way to begin work at that project, by the way. There is also no major donor anywhere to be found for that work. Instead, any funds that seem to be appearing get attacked and sucked into projects that miss the point, being controlled by folks who oppose openness (e.g. WISPs wanting exclusive ownership of a market, such as so called SuperWiFi or whitespaces). I did once come close to a useful award when I was at MIT Media Lab, from NSF. But after the award, the funding was cut by 90%, leaving just enough to support a Master's thesis on co-channel sharing, using two 1st Gen USRPs. Using my own funds, spare time, and bubblegum and baling wire, I've slowly begun work on extra wideband FPGA based sounding-centric sharing in the 10 GHz Ham band. (500 MHz wide modulation), where I can self certify multiple stations in a network.
>
> But the point is, I've failed, because there is less than zero support. There is active opposition, on top of cluelessness.
>
> Paul Baran tried in 1993 to push forward a similar agenda, famously. 99% of his concepts died.

Cite?

One of the things that bothers me about packet processing is that
Donald Davies (the oft uncredited other founder of the concept) wrote
11 volumes on this subject. So far as I know, those have vanished to
history.

Periodically, when I get stuck on something in this field, I fantasize
that scribbled in the margin of volume 9 was the solution to the
problem.

> Thanks to Apple, and lots of others, we got WiFi, barely. Industry hated that, and vow never to let that ever happen again.

It really was a strange convolution of circumstances that led to wifi.
When i first got it working in 1998, metricom ruled the world. They
failed. After that, nobody thought it was feasible at scale until the
concept of a mac retry emerged to fix the packet loss problem, and APs
to provide a central clock (best we could do with the DSPs then).  So
a window emerged (and yes, hugely driven by apple, but also by huge
popular demand for "wireless freedom") to put "buggy" wireless tech on
the crap 2.4 band in the hands of the people, it got established and
made the coffee shop a workplace, and bigcos attempting to wipe it out
(and largely, in the last few years, succeeding in dislodging it) have
had an uphill battle.

If metricom had succeeded, or the celluar folk got their
implementations working only a few years faster, it would be a very
different world.

(this history is all covered in my MIT preso here:
https://www.youtube.com/watch?v=Wksh2DPHCDI&t=2007s - david was at
that one)

>
> So Dave, I salute you and Toke and the others. I salute Tim Shepard, who also moved the ball in his PhD thesis, only to hit the same wall of opposition.

Soooo many others involved, felix feitkau in particular comes to mind.

Still, I think fixing the "wifi anomaly" is the greatest achievement
of my career... and toke's hasn't even officially started yet! Someday
perhaps that will be worth a medal, or an small entry for us in
wikipedia.

> It's so sad. We get shit like the "Obama band" proposed by PCAST, and are told to be thankful.

Let's not get started on that or whitespaces today.

> UWB failed miserably, too.

I wish that could be resurrected.

> My advice to any young smart innovator: don't touch wireless unless you are working for an incumbent. Expect the incumbents and governments to close and destroy wireless innovation.

I agree. Well, I do have some hope and interest in spacex's
constellation, but I remember teledesic's failure too well.

>
> Really. You will be in a world of hurt, and NO ONE will support anything. Not even VCs.
>
> Very sorry to say this. I had hoped Make WiFi Fast would have gone somewhere. I mourn its passing.

It's not dead, I'm just closing my lab.

In the document I cited for more wifi fixes, things like dynamically
scaling down the announced txop under contention, lowering retries,
offering a little less protection for packets when overloaded, a "tx
is almost done" interrupt, etc, are all things I expect vendors and
open source folk to try. I keep hoping minstrel-blues will land. Etc.
Outside the US there's still a lot of positive activity. Products like
eero and google wifi continue to sell like hotcakes, as well as tons
of cheaper gear, and iot, etc, etc.

And there's some good progress in 802.11ax.

fq_codel for wifi + these mods will make wifi continue to be more than
competitive with the upcoming 5G stuff.

But i don't need to be the one to implement or test them. I should
have shut things down when the shuttleworth grant didn't come through.
When you can no longer get up in the morning to work, nor able to hold
the wifi standard and related code in your head, it's time to move on.

It is bothersome to me that the ISPs don't seem to realize that their
business will fail unless they have good wifi, but the big ones are
out there merging with the LTE folk and don't care either.

Wifi's had a great run. I think here - with 5 years of work - we've
extended its life another 5 years - at least. Still, unless
applications emerge again that need good low latency (like vr) over
wifi, nothings going to drive those down further to compete with 5g.

>
> -----Original Message-----
> From: "Dave Taht" <dave.taht at gmail.com>
> Sent: Fri, Aug 24, 2018 at 4:10 pm
> To: bloat-announce at lists.bufferbloat.net, "bloat" <bloat at lists.bufferbloat.net>, "Make-Wifi-fast" <make-wifi-fast at lists.bufferbloat.net>, cerowrt-devel at lists.bufferbloat.net
> Cc: bloat-announce at lists.bufferbloat.net, "bloat" <bloat at lists.bufferbloat.net>, "Make-Wifi-fast" <make-wifi-fast at lists.bufferbloat.net>, cerowrt-devel at lists.bufferbloat.net
> Subject: [Cerowrt-devel] closing up my make-wifi-fast lab
>
> All:
>
> It is with some regret that I am announcing the closing of my
> make-wifi-fast lab at the end of this month.
>
> Over the years we have relied on the donation of lab space from
> ISC.org, georgia tech, the LINCs, and the University of Karstadt and
> elsewhere - but my main base of operation has always been the
> "yurtlab", in a campground deep in the los gatos hills where I could
> both experiment and deploy wifi fixes[0] at scale. CeroWrt, in
> particular, was made here.
>
> During the peak of the make-wifi-fast effort I rented additional space
> on the same site, which at peak had over 30 routers in a crowded
> space, competing. Which I (foolishly) kept, despite the additional
> expense. Having heat in the winter and aircond in the summer was
> helpful.
>
> With ongoing donations running at $90/month[1] - which doesn't even
> cover bufferbloat.net's servers in the cloud - my biggest expense has
> been keeping the lab at lupin open at $1800/mo.
>
> I kept the lab going through the sch_cake and openwrt 18.06 release
> process, and I'm now several months behind on rent[3], and given how
> things have gone for the past 2 years I don't see much use for it in
> the future. Keeping it open, heated and dry in the winter has always
> been a problem also. I'm also aware of a few larger, much better
> equipped wifi labs that have thoroughly tested our "fq_codel for
> wifi"[4] work that finally ends the "wifi performance anomaly". it's
> in multiple commercial products now, we're seeing airtime fairness
> being actually *marketed* as a wifi feature, and I kind of expect
> deployment be universal across all mediatek mt76, and qualcomm ath9k
> and ath10k based products in the next year or two. We won, big, on
> wifi. Knocked it out of the park. Thanks all!
>
> Despite identifying all kinds of other work[5] that can be done to
> make wifi better, no major (or even minor) direct sponsor has ever
> emerged[2] for the make-wifi-fast project. We had a small grant from
> comcast, a bit of support from nlnet also, I subsidized what I did
> here from other work sources, toke had his PHD support, and all the
> wonderful volunteers here... and that's it.
>
> Without me being able, also, to hire someone to keep the lab going, as
> I freely admit to burnout and PTSD on perpetually reflashing and
> reconfiguring routers...
>
> I'm closing up shop here to gather enough energy, finances, and time
> for the next project, whatever it is.
>
> The make-wifi-fast mailing list and project will continue, efforts to
> make more generic the new API also, and hopefully there's enough users
> out there to
> keep it all going forward without the kind of comprehensive testing I
> used to do here.
>
> If anyone feels like reflashing, oh, 30 bricked routers of 8 different
> models, from serial ports (in multiple cases, like the 6 uap-ac-lites,
> via soldiering on headers), I'll gladly toss all the extra equipment
> in the lab in a big box and ship them to you. Suggestions for a
> suitable donation target are also of interest.
>
> The yurtlab has been an amazing, totally unique, unusual (and
> sometimes embarrassing [6]) place to work and think, but it's time to
> go.
>
> Perhaps I'll convince my amazingly supportive landlord to let me leave
> behind a plaque:
>
> "On this spot bufferbloat on the internet and in WiFi was fixed, 2011-2018".
>
> Sincerely,
> Dave Taht
>
> [0] https://lwn.net/Articles/705884/ "How we made wifi fast again"
> [1] https://www.patreon.com/dtaht
> [2] Like adrian chadd's infamous flameout - I too, give up on wifi.
> There's gotta be some other tech worth working on. What we shipped is
> "good enough" to carry a few years though.
> [3] This is not a passive-aggressive request for help making rent next
> month, given all the other problems I have, it's best to close up shop
> while I look for a new gig.
> [4] https://arxiv.org/pdf/1703.00064.pdf "ending the wifi anomaly"
> [5] https://docs.google.com/document/d/1Se36svYE1Uzpppe1HWnEyat_sAGghB3kE285LElJBW4/edit#
> [6] https://www.cringely.com/2012/10/01/clothing-may-be-optional-but-bufferbloat-isnt/
> _______________________________________________
> Cerowrt-devel mailing list
> Cerowrt-devel at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cerowrt-devel
>
>


-- 

Dave Täht
CEO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-669-226-2619



More information about the Bloat mailing list