From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-wm1-x32b.google.com (mail-wm1-x32b.google.com [IPv6:2a00:1450:4864:20::32b]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by lists.bufferbloat.net (Postfix) with ESMTPS id C9B823B2A4; Wed, 20 Apr 2022 06:29:42 -0400 (EDT) Received: by mail-wm1-x32b.google.com with SMTP id p189so886345wmp.3; Wed, 20 Apr 2022 03:29:42 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20210112; h=mime-version:references:in-reply-to:from:date:message-id:subject:to :cc; bh=5QzoN+1d5zYg52zIOxVtfL1vJnk7RrPXsNSLCmI74Z4=; b=B8zXv9ssgiHC4DazfayvgzRiTZVxOzGnRuQqsSIzNPxRb90udLfNX0/S6wBjPNrxAM xoY11zfl3DTGYjLYLOTO5sZMzI5AGaljGQY0QpPDLEF2DuGtJnSeLglJA7t37JNvluSF TD/irNAu9N6x8n06wLg56bymDqFcwicgCQb5ed5RMfWqncxXkjQ13Jvcz1rE2deN8WVe L4hQsiaf8J0YFwcwHAyoOhd0FiDr3OugQvX93K6z2FxVa7yfN/UnDnYvW+jS94w0F4yn Ndyg78cuQwX1YmgDcfIQNldtKPAMn0lV+PpljPIUk918XYZccAbRQ1Ls8lm92xnFyMZR yCEQ== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:mime-version:references:in-reply-to:from:date :message-id:subject:to:cc; bh=5QzoN+1d5zYg52zIOxVtfL1vJnk7RrPXsNSLCmI74Z4=; b=g6M2gMGbTgMMXOFoEpHwju6JNnvAeJaJ4kLGxBmeHKuqKxjZwR1lIwG1wygpQCHWAf 6Q/pAByKLaJpEre/Y7+Jr4B+ax8rh2ROSoxgpNj9JFXpv/U/wdZHJIkyZ18ytC97yNI/ Gqiz8ecQ6lWa8rOjMseQ37c3hUAIumfGLiI7UIoV6wGvhLH2a5OFjCDs30ORUWofiSr/ 1cX00KEKjwqsP/O0bM94YOETBkj912iOo+JNrrpoKqCDPyJupqwBZ7dVWlf0QmtVX2wp Ar3FIcz0WIWXiOA/gYxYpOkCHVkmDUT4Sm3rC8GESJNridecYm9LcF6109ROsbhbAMKq vZkQ== X-Gm-Message-State: AOAM530+8G03nDdi0PautoQeHl4No8FzA8ulr3RvE6k2oqGlAch+6doZ IwOYa8xSxwfvyXTdCapUInKx7oHIEeQ1V4neFhQ= X-Google-Smtp-Source: ABdhPJzfCwsZpNqZ4wUr0v1J0l3kYFXU1kivDHZ0ceWvSKD4irBRN4WuZH8NxHhth+qY1BLDbzCdCNgTSQpeHYD5B5I= X-Received: by 2002:a05:600c:1d8b:b0:392:ab4f:365d with SMTP id p11-20020a05600c1d8b00b00392ab4f365dmr2369450wms.113.1650450581653; Wed, 20 Apr 2022 03:29:41 -0700 (PDT) MIME-Version: 1.0 References: In-Reply-To: From: Andrew Somerville Date: Wed, 20 Apr 2022 11:29:06 +0100 Message-ID: To: "Luis A. Cornejo" Cc: =?UTF-8?Q?Joel_Wir=C4=81mu_Pauling?= , cerowrt-devel , bloat Content-Type: multipart/alternative; boundary="000000000000a38c5305dd1378ae" X-Mailman-Approved-At: Sat, 23 Apr 2022 15:08:58 -0400 Subject: Re: [Cerowrt-devel] [Bloat] android, sqm-autorate and lte and videoconferencing X-BeenThere: cerowrt-devel@lists.bufferbloat.net X-Mailman-Version: 2.1.20 Precedence: list List-Id: Development issues regarding the cerowrt test router project List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 20 Apr 2022 10:29:43 -0000 --000000000000a38c5305dd1378ae Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Dear Joel, Luis, Yes exactly; the bash CAKE-autorate implementation seeks to increase bandwidth on load subject to the RTT not spiking. I think of it a little like a 'turbo' function on a CPU. It allows supervised excursions beyond the base bandwidth (safe harbour) for periods of heavy use, and otherwise returns to the safe harbour. Such supervised excursions are helpful to allow downloads to finish quicker and without hurting videoconferencing, and the safe harbour is helpful for general use when higher CAKE bandwidth is not needed anyway. If even the safe harbour causes trouble (e.g. a tsunami), it will keep dialing down to the set minimum bandwidth as required. This is the thinking behind having the set minimum, base and maximum for both download an dupload. Whilst the approach seems to be working pretty well on LTE and Starlink variable connections (I mean for sure it beats just using CAKE with a fixed bandwidth on my LTE, and there are a few pretty positive reports from others users), I imagine that it will not help with CPU saturation. I read a lot of reports about CAKE eating up too many cycles (e.g. on the RT3200 CAKE tops out at 500Mbit/s), so it seems there is a lot of appetite for making CAKE more efficient. But this is not my area of expertise. Kind regards, Andrew On Sun, 17 Apr 2022 at 06:28, Luis A. Cornejo wrote: > It still is Cake underneath doing the work, autorate is just adjusting > available bandwidth to get the best latency at the best possible bandwidt= h. > I would think it's the same limitation. I believe Dave Taht mentioned > possibly making cake take advantage of multiple cores in the next > version/rewrite of Cake. > > On Fri, Apr 15, 2022 at 4:41 PM Joel Wir=C4=81mu Pauling > wrote: > >> Whilst I understand it's not designed for High Bandwidth uplinks - how >> does it scale to 1 to 10Gbit symmetrical (or near symmetrical) uplinks ? >> >> One of the problems i've had with Cake is that it becomes CPU bound >> beyond around 3 or 4 gigabit. >> >> On Sat, 16 Apr 2022 at 07:41, Andrew Somerville >> wrote: >> >>> Dear all, >>> >>> Thank you very much for the introduction Sebastian. >>> >>> To give some context, my wife and I relocated to the Scottish Highlands >>> and now rely upon a 4G LTE connection for work and personal use through >>> Vodafone UK. I have spent quite a lot of time working on this autorate >>> problem and have tried to leverage Sebastian's expertise in this field = as >>> much as possible. I have tried to keep it as simple as possible with so= me >>> rationale and objective criticism behind the major logic. I value feedb= ack >>> and criticism. >>> >>> I now use the bash implementation in my main branch on my RT3200 router >>> as a service 24/7. I have rewritten it a few times and will do so again= , or >>> switch to another better approach if available. There have been a few p= orts >>> of some of the earlier versions like this Golang version ( >>> https://github.com/notsure2/cake-autorate). >>> >>> Kind regards, >>> >>> Andrew >>> >>> >>> On Wed, 6 Apr 2022 at 22:42, Dave Taht wrote: >>> >>>> it's looking promising. >>>> >>>> in trying to get an android to do better this recent ML paper crossed >>>> my desk: >>>> >>>> https://arxiv.org/pdf/2007.02735.pdf >>>> >>>> On Wed, Apr 6, 2022 at 1:38 PM Sebastian Moeller >>>> wrote: >>>> > >>>> > Dear Dave, dear all >>>> > >>>> > please, let me introduce Andrew to this list, who is the driving >>>> force behind CAKE-autorate's design and implementation (which started = from >>>> a more theoretical discussion in the OpenWrt forum before turning into >>>> something tangible). There are other alternative approaches for the >>>> rate-tracking problem many discussed in this longish forum thread: >>>> https://forum.openwrt.org/t/cake-w-adaptive-bandwidth/108848 (which is >>>> great as this occasionally leads to quite interesting discussion about= how >>>> the different teams tackle common issues) but Andrew's autorate appear= s to >>>> the fastest moving with low software requirements (every router should= run >>>> bash anyway ;) ). >>>> > >>>> > Kind Regards >>>> > Sebastian >>>> > >>>> > >>>> > > On Apr 6, 2022, at 17:43, Dave Taht wrote: >>>> > > >>>> > > For the past several days, I have been very successfully using >>>> > > variants of the cake-autorate code to manage my connections on the >>>> > > boat, for which I use a tether to my laptop. >>>> > > >>>> > > https://github.com/lynxthecat/CAKE-autorate >>>> > > >>>> > > Although this test claims my link was inadequate for a good >>>> videoconference >>>> > > >>>> > > >>>> https://www.waveform.com/tools/bufferbloat?test-id=3D964831e5-30f9-469= 5-bfbd-b58da0a759f3 >>>> > > >>>> > > they have all been perfect (and that test was conducted during an >>>> > > actual zoom conference). The code does not grab as much bandwidth = as >>>> > > it could, when available, but I'll settle for perfect >>>> > > videoconferencing. >>>> > > >>>> > > Anyway... what I used to do was attach the phone to a router share= d >>>> > > boat-wide that did this stuff, but it would be nice to move the >>>> > > algorithm directly into an android. My hope is that more modern >>>> > > androids are running a recent enough kernel(?) to have cake, but >>>> it's >>>> > > been a long time since I built anything for android, and am >>>> wondering >>>> > > if there is a lte/5g tablet or phone or dedicated lte router "out >>>> > > there" that can be hacked on? >>>> > > >>>> > > >>>> > > >>>> > > -- >>>> > > I tried to build a better future, a few times: >>>> > > https://wayforward.archive.org/?site=3Dhttps%3A%2F%2Fwww.icei.org >>>> > > >>>> > > Dave T=C3=A4ht CEO, TekLibre, LLC >>>> > > _______________________________________________ >>>> > > Cerowrt-devel mailing list >>>> > > Cerowrt-devel@lists.bufferbloat.net >>>> > > https://lists.bufferbloat.net/listinfo/cerowrt-devel >>>> > >>>> >>>> >>>> -- >>>> I tried to build a better future, a few times: >>>> https://wayforward.archive.org/?site=3Dhttps%3A%2F%2Fwww.icei.org >>>> >>>> Dave T=C3=A4ht CEO, TekLibre, LLC >>>> >>> _______________________________________________ >>> Cerowrt-devel mailing list >>> Cerowrt-devel@lists.bufferbloat.net >>> https://lists.bufferbloat.net/listinfo/cerowrt-devel >>> >> _______________________________________________ >> Bloat mailing list >> Bloat@lists.bufferbloat.net >> https://lists.bufferbloat.net/listinfo/bloat >> > --000000000000a38c5305dd1378ae Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable
Dear Joel, Luis,

Yes exactly; the bash = CAKE-autorate implementation seeks to increase bandwidth on load subject to= the RTT not spiking.=C2=A0

I think of it a little= like a 'turbo' function on a CPU. It allows supervised excursions = beyond the base bandwidth (safe harbour) for periods of heavy use, and othe= rwise returns to the safe harbour. Such supervised excursions are helpful t= o allow downloads to finish quicker and without hurting videoconferencing, = and the safe harbour is helpful for general use when higher CAKE bandwidth = is not needed anyway. If even the safe harbour causes trouble (e.g. a tsuna= mi), it will keep dialing down to the set minimum bandwidth as required. Th= is is the thinking behind having the set minimum, base and maximum for both= download an dupload.

Whilst the approach seems to= be working pretty well on LTE and Starlink variable connections (I mean fo= r sure it beats just using CAKE with a fixed bandwidth on my LTE, and there= are a few pretty positive reports from others users), I imagine that it wi= ll not help with CPU saturation.=C2=A0

I read a lo= t of reports about CAKE eating up too many cycles (e.g. on the RT3200 CAKE = tops out at 500Mbit/s), so it seems there is a lot of appetite for making C= AKE more efficient. But this is not my area of expertise.

Kind regards,

Andrew

On Sun, 17 Apr 2= 022 at 06:28, Luis A. Cornejo <luis.a.cornejo@gmail.com> wrote:
It still is Cake underneath doi= ng the work, autorate is just adjusting available bandwidth to get the best= latency at the best possible bandwidth. I would think it's the same li= mitation. I believe Dave Taht mentioned possibly making cake take advantage= of multiple cores in the next version/rewrite of Cake.

On Fri, Apr 15, = 2022 at 4:41 PM Joel Wir=C4=81mu Pauling <joel@aenertia.net> wrote:
Whilst I understand i= t's not designed for High Bandwidth uplinks - how does it scale to 1 to= 10Gbit symmetrical (or near symmetrical) uplinks ?

One of the problem= s i've had with Cake is that it becomes CPU bound beyond around 3 or 4 = gigabit.

On Sat, 16 Apr 2022 at 07:41, Andrew Somerville <aesomerville@gmail.c= om> wrote:
Dear all,

Thank you very much for the= introduction Sebastian.=C2=A0

To give some contex= t, my wife and I relocated to the Scottish Highlands and now rely upon a 4G= LTE connection for work and personal use through Vodafone UK. I have spent= quite a lot of time working on this autorate problem and have tried to lev= erage Sebastian's expertise in this field as much as possible. I have t= ried to keep it as simple as possible with some rationale and objective cri= ticism behind the major logic. I value feedback and criticism.=C2=A0
<= div>
I now use the bash implementation in my main branch on m= y RT3200 router as a service 24/7. I have rewritten it a few times and will= do so again, or switch to another better approach if available. There have= been a few ports of some of the earlier versions like this Golang version = (ht= tps://github.com/notsure2/cake-autorate).=C2=A0

Kind regards,=C2=A0

Andrew

<= /div>
O= n Wed, 6 Apr 2022 at 22:42, Dave Taht <dave.taht@gmail.com> wrote:
it's looking promising.

in trying to get an android to do better this recent ML paper crossed my de= sk:

https://arxiv.org/pdf/2007.02735.pdf

On Wed, Apr 6, 2022 at 1:38 PM Sebastian Moeller <moeller0@gmx.de> wrote:
>
> Dear Dave, dear all
>
> please, let me introduce Andrew to this list, who is the driving force= behind CAKE-autorate's design and implementation (which started from a= more theoretical discussion in the OpenWrt forum before turning into somet= hing tangible). There are other alternative approaches for the rate-trackin= g problem many discussed in this longish forum thread: https://forum.openwrt.org/t/cake-w-adaptive-bandwidth/108848= (which is great as this occasionally leads to quite interesting discus= sion about how the different teams tackle common issues) but Andrew's a= utorate appears to the fastest moving with low software requirements (every= router should run bash anyway ;) ).
>
> Kind Regards
>=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0Sebastian
>
>
> > On Apr 6, 2022, at 17:43, Dave Taht <dave.taht@gmail.com> wrote:
> >
> > For the past several days,=C2=A0 I have been very successfully us= ing
> > variants of the cake-autorate code to manage my connections on th= e
> > boat, for which I use a tether to my laptop.
> >
> > https://github.com/lynxthecat/CAKE-autorate=
> >
> > Although this test claims my link was inadequate for a good video= conference
> >
> > h= ttps://www.waveform.com/tools/bufferbloat?test-id=3D964831e5-30f9-4695-bfbd= -b58da0a759f3
> >
> > they have all been perfect (and that test was conducted during an=
> > actual zoom conference). The code does not grab as much bandwidth= as
> > it could, when available, but I'll settle for perfect
> > videoconferencing.
> >
> > Anyway... what I used to do was attach the phone to a router shar= ed
> > boat-wide that did this stuff, but it would be nice to move the > > algorithm directly into an android. My hope is that more modern > > androids are running a recent enough kernel(?) to have cake, but = it's
> > been a long time since I built anything for android, and am wonde= ring
> > if there is a lte/5g tablet or phone or dedicated lte router &quo= t;out
> > there" that can be hacked on?
> >
> >
> >
> > --
> > I tried to build a better future, a few times:
> > https://wayforward.archiv= e.org/?site=3Dhttps%3A%2F%2Fwww.icei.org
> >
> > Dave T=C3=A4ht CEO, TekLibre, LLC
> > _______________________________________________
> > Cerowrt-devel mailing list
> > Cerowrt-devel@lists.bufferbloat.net
> > https://lists.bufferbloat.net/listinfo= /cerowrt-devel
>


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

Dave T=C3=A4ht CEO, TekLibre, LLC
_______________________________________________
Cerowrt-devel mailing list
Ce= rowrt-devel@lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/cerowrt-d= evel
_______________________________________________
Bloat mailing list
Bloat@list= s.bufferbloat.net
https://lists.bufferbloat.net/listinfo/bloat
--000000000000a38c5305dd1378ae--