From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from vsmx001.dclux.xion.oxcs.net (vsmx001.dclux.xion.oxcs.net [185.74.65.81]) (using TLSv1.2 with cipher ADH-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by lists.bufferbloat.net (Postfix) with ESMTPS id C98AA3CB37 for ; Tue, 30 Aug 2022 16:09:37 -0400 (EDT) Received: from proxy-2.proxy.oxio.ns.xion.oxcs.net (proxy-2.proxy.oxio.ns.xion.oxcs.net [83.61.18.4]) by mx-out.dclux.xion.oxcs.net (Postfix) with ESMTPA id A429B8C030C for ; Tue, 30 Aug 2022 20:09:35 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=simple/simple; d=dclux.xion.oxcs.net; s=mail1; t=1661890176; bh=aSlgT04JcrEr4vTDygYAOUU+jrwoL9H97oL/9bI92n0=; h=Date:From:To:In-Reply-To:References:Subject:From; b=jwiMvNsgG+hzlXQbx6syUhHMiQvGjg6eGktpPZkV0+4UzjoUCK6fic3dq1EjDm5LX 53IE/CjSYMOteMufUP59j1tKKtU4J99f3MIXRnaFXnwKX/2OGqqPNTRkHC5oY7MIVI Noq2N+6dyfwc2GLJBpLzs3BU0fLe64RNqFJHwyf0agKLpYZ0arfiUcj+SsU3WVqRer UIPnKbEMBrV3oUrLu20tJpPkd1e05Ki0I7DhW6lXfaV4TKvKjcKhfwfpJ46BGj0zn4 JLcO3dpBLbWRecPYd2KONarggq8LuzJezijF/c4SIhyKmGG2rT4gFe3UrvNLyfba25 VjzjFEzK4oGuQ== Date: Tue, 30 Aug 2022 22:09:28 +0200 From: Mike Puchol To: starlink@lists.bufferbloat.net Message-ID: <7b7bff13-b0ef-41ad-9a40-d12bfe5a0dfe@Spark> In-Reply-To: References: <1661878433.14064713@apps.rackspace.com> X-Readdle-Message-ID: 7b7bff13-b0ef-41ad-9a40-d12bfe5a0dfe@Spark MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="630e6e7e_26f324ba_2eb" X-VadeSecure-Status: LEGIT X-VADE-STATUS: LEGIT Subject: Re: [Starlink] Starlink "beam spread" X-BeenThere: starlink@lists.bufferbloat.net X-Mailman-Version: 2.1.20 Precedence: list List-Id: "Starlink has bufferbloat. Bad." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 30 Aug 2022 20:09:38 -0000 --630e6e7e_26f324ba_2eb Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline Given that I may have triggered this, and with my limited understanding o= f how Starlink is put together from a resource allocation standpoint, fir= st, some caveats: =E2=80=A2 A simulation without inside knowledge is always going to be an = approximation. In my case, it=E2=80=99s usually the best-case scenario. =E2=80=A2 The simulation of a country should not be done in isolation to = increase realism. Many of the satellites over the northern US are also se= rving Canada, which means there are less resources available to US cells = than the simulation portrays. =E2=80=A2 The real constellation would allocate resources based on demand= , primarily, and Starlink knows said demand quite accurately (once ESIMs = are in place, this will change=21). Thus, cells which are in very densely= populated areas, which I include, would likely receive very few resource= s, if any. Many empty cells or those with no customers won=E2=80=99t rece= ive resources either. Starlink has essentially these mechanisms to increase the number of cells= covered by the constellation at any given time (others like number of be= ams are fixed in hardware): =E2=80=A2 TDM aka beamhopping: split 100% of =E2=80=9Cairtime=E2=80=9D in= to frames, and allocate certain number of frames to more than one cell. C= an provide asymmetric capacity based on demand per cell. The switching ti= me from cell to cell is the driver for how far you can split the beam in = the time domain, before the switching overhead starts killing performance= . =E2=80=A2 Beam spread: as the beam is steered away from nadir, it becomes= larger and elliptical, thus covering more than one cell (apart from the = target cell). Allows to split the beam resources allocated to the center = cell amongst all other cells in the =46OR (field of regard). Cannot be as= ymmetric without enforcing terminal discipline, where each terminal is re= ceiving 100% of the downlink (which could be 10% of the full beam if TDM = splits into 10 cells), but is only grabbing what is addressed to it speci= fically (akin to an LTE network). =E2=80=A2 Multiple beams over a single cell: up to eight spot beams can b= e projected onto a single cell without running into EP=46D limits, as lon= g as each one uses one of the eight frequencies available. These simultan= eous beams could come from one or more satellites. This is how you can ge= t additional capacity to a cell, for example, to compensate for reduction= by TDM. Two beams at 50% duty cycle make up for one full beam. The advan= tage is spatial diversity, where a terminal that has one satellite obstru= cted could opt from a beam from a different, non-obstructed satellite. We must also understand provisioned vs. advertised capacity. In Kenya, we= provision 1.5 Mbps per customer, but advertise and sell a 5 Mbps service= . Actual average per customer is 1.2 Mbps. If we took a single beam capab= le of 700 Mbps (many observations and circumstantial evidence support thi= s figure), at 1.5 Mbps provisioned it would serve =7E466 customers. Once = you start applying TDM, beam spread, etc. the number reduces, and can onl= y be compensated by adding additional beams. It=E2=80=99d be interesting = to hear what ISPs in the US provision their customers with. Of course fiber is the top option if you can get it economically, but if = there are vast regions of the US (not considered a developing country) st= ill not serviced by it, or even by fast WISPs, then there are reasons to = look for alternatives - I=E2=80=99m willing to bet a financial analysis d= oesn=E2=80=99t warrant laying thousands of kilometers of fiber to serve r= elatively few customers. Starlink is providing significant service levels= to people who could only dream, with all its growing pains and inefficie= ncies. If you start talking about =E2=80=9Cconnecting the unconnected=E2=80=9D (= a kitten dies every time that one is said), then fiber is one option, but= it becomes more relevant for major backhaul, not even middle or last mil= e. People who have a disposable income of =241 to =245 per month just can= not be serviced by a financially viable service that relies on fiber. Her= e, you need to start getting creative. Best, Mike On Aug 30, 2022, 19:32 +0200, Doc Searls via Starlink , wrote: > All good points. > > I'm also wondering if (and how) Starlink is improving any satellite gea= r in successive launches. And, if that's the case, what would be the uppe= r limit to what's possible with the system=3F > > I ask the first question because Starlink has been deorbiting quite a f= ew satellites... > > https://spacenews.com/spacex-launches-starlink-satellites-as-it-deorbit= s-original-ones/ > > https://www.space.com/spacex-starlink-satellite-deorbit-video > > ... while launching many new ones. > > =46or example, there will be a =46alcon launch from Vandenberg, of seve= ral dozen satellites, at 10:30 (or :40) PM Pacific time on Wednesday nigh= t (though there are conflicting reports, and launches often get canceled)= : > > https://www.edhat.com/news/spacex-starlink-launch-rescheduled-for-tuesd= ay > > https://www.spacelaunchschedule.com/category/vandenberg-sfb/ > > https://www.space.com/spacex-starlink-group-3-4-launch-rocket-landing > > https://www.ksby.com/news/local-news/spacex-targets-tuesday-night-for-f= alcon-9-launch-of-starlink-satellites > > A late evening launch time makes for good viewing because it's dark eno= ugh to see the launch from a distance, and the rocket hits sunlight at th= e edge of space, where exhaust moves outward in all directions uncontaine= d by atmosphere, leaving a tubular trail in the sky. > > Here is a collection of screen grabs from a camcorder recording of a la= unch in 2005: https://www.flickr.com/photos/docsearls/albums/999576 This = launch will be later in the evening, but still quite visible. One big dif= ference will be the return trip of the first stage to a platform out in t= he ocean. I caught one of those in this series of shots here: https://www= .flickr.com/photos/docsearls/albums/72157701027229232 > > (=46orgive my indulgence in space-freakery. I do enjoy this stuff, and = I'm not here in Santa Barbara often enough. But I am here now, so I'll be= shooting it again, this time with a new camera and a longer lens.) > > Doc > > > On Aug 30, 2022, at 9:53 AM, David P. Reed via Starlink wrote: > > > > I have no clue why this matters (other than this is in color). > > > > The phased array antennas used by Starlink are quite limited - in par= ticular, there are 4 on each satellite and each earth-ground path is half= -duplex, TDM, essentially. Limited by hardware. The problem of signal equ= alization and quantization limits prevent =22space division multiplexing=22= and =22frequency division multiplexing=22 in practice. > > > > The 4 msec =22turnaround time=22 at the physical level (satellite) me= ans that time from a packet arriving at one end to be sent to the other e= nd of the sat-dishy links gets worse the more dishys are served by one of= the 4 antennas on the satellite. > > > > trying to increase the coverage of an individual satellite basically = means serving more dishys per satellite, with less total bit rate, and mu= ch longer latency due to the half duplexness. > > > > Now if the total bit rate of a sat-to-dishy link were, say, 1 Gigabit= , like an 802.11ac AP gives you, and the turnaround time were under 1 mic= rosecond rather than 4 msec. maybe then you could get reasonable Internet= service to dishys. > > > > But 240 Mb/s or 172 Mb/s as proposed for getting a bit more coverage = per satellite=3F This is nowhere near competitive with what we expect in = the US. > > > > Sorry to rain on all the techy dreaming. > > > > =46irst, it's worth looking at all the problems currently in Wi=46i p= erformance when you share an AP with multiple active stations using 100's= of Gb/s on the average (not just occasionally). > > > > Dave - you tried in =22make-wifi-fast=22, and the architecture gets i= n the way there. (yeah you can get point to point gigabit/sec single file= transfers, but to do that you invoke features that destroy latency and i= ntroduce huge variability if you share the AP at all, for these reasons).= > > > > Starlink is a good =22last resort=22 service as constituted. But fibe= r and last few-hundred meters wireless is SO much better able to deliver = good Internet service scalably. > > Even that assumes fixing the bufferbloat that the Starlink folks don'= t seem to be able to address... > > =5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F= =5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F > > Starlink mailing list > > Starlink=40lists.bufferbloat.net > > https://lists.bufferbloat.net/listinfo/starlink > > =5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F= =5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F > Starlink mailing list > Starlink=40lists.bufferbloat.net > https://lists.bufferbloat.net/listinfo/starlink --630e6e7e_26f324ba_2eb Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline
Given that I may have triggered this, and with my l= imited understanding of how Starlink is put together from a resource allo= cation standpoint, first, some caveats:
  • A simulation without inside knowledge is always going to be an approx= imation. In my case, it=E2=80=99s usually the best-case scenario.
  • The simulation of a country should not be done in isolation to increa= se realism. Many of the satellites over the northern US are also serving = Canada, which means there are less resources available to US cells than t= he simulation portrays.
  • The real constellation would allocate resources based on demand, prim= arily, and Starlink knows said demand quite accurately (once ESIMs are in= place, this will change=21). Thus, cells which are in very densely popul= ated areas, which I include, would likely receive very few resources, if = any. Many empty cells or those with no customers won=E2=80=99t receive re= sources either.&=23160;
Starlink has essentially these mechanisms to increa= se the number of cells covered by the constellation at any given time (ot= hers like number of beams are fixed in hardware):
  • TDM aka beamhopping: split 100% of =E2=80=9Cairtime=E2=80=9D into fra= mes, and allocate certain number of frames to more than one cell. Can pro= vide asymmetric capacity based on demand per cell. The switching time fro= m cell to cell is the driver for how far you can split the beam in the ti= me domain, before the switching overhead starts killing performance.

  • Beam spread: as the beam is steered away from nadir, it becomes large= r and elliptical, thus covering more than one cell (apart from the target= cell). Allows to split the beam resources allocated to the center cell a= mongst all other cells in the =46OR (field of regard). Cannot be asymmetr= ic without enforcing terminal discipline, where each terminal is receivin= g 100% of the downlink (which could be 10% of the full beam if TDM splits= into 10 cells), but is only grabbing what is addressed to it specificall= y (akin to an LTE network).

  • Multiple beams over a single cell: up to eight spot beams can be proj= ected onto a single cell without running into EP=46D limits, as long as e= ach one uses one of the eight frequencies available. These simultaneous b= eams could come from one or more satellites. This is how you can get addi= tional capacity to a cell, for example, to compensate for reduction by TD= M. Two beams at 50% duty cycle make up for one full beam. The advantage i= s spatial diversity, where a terminal that has one satellite obstructed c= ould opt from a beam from a different, non-obstructed satellite.
We must also understand provisioned vs. advertised = capacity. In Kenya, we provision 1.5 Mbps per customer, but advertise and= sell a 5 Mbps service. Actual average per customer is 1.2 Mbps. If we to= ok a single beam capable of 700 Mbps (many observations and circumstantia= l evidence support this figure), at 1.5 Mbps provisioned it would serve =7E= 466 customers. Once you start applying TDM, beam spread, etc. the number = reduces, and can only be compensated by adding additional beams. It=E2=80= =99d be interesting to hear what ISPs in the US provision their customers= with.

Of course fiber is the top option if you can get it economically, but if = there are vast regions of the US (not considered a developing country) st= ill not serviced by it, or even by fast WISPs, then there are reasons to = look for alternatives - I=E2=80=99m willing to bet a financial analysis d= oesn=E2=80=99t warrant laying thousands of kilometers of fiber to serve r= elatively few customers. Starlink is providing significant service levels= to people who could only dream, with all its growing pains and inefficie= ncies.

If you start talking about =E2=80=9Cconnecting the unconnected=E2=80=9D (= a kitten dies every time that one is said), then fiber is one option, but= it becomes more relevant for major backhaul, not even middle or last mil= e. People who have a disposable income of =241 to =245 per month just can= not be serviced by a financially viable service that relies on fiber. Her= e, you need to start getting creative.

Best,

Mike
On Aug 30, 2022, 19:32 +0200, Doc S= earls via Starlink <starlink=40lists.bufferbloat.net>, wrote:
=
All good points.

I'm also wondering if (and how) Starlink is improving any satellite gear = in successive launches. And, if that's the case, what would be the upper = limit to what's possible with the system=3F

I ask the first question because Starlink has been deorbiting quite a few= satellites...

https://spacenews.com/spacex-launches-starlink-satellites-as-it-deorbits-= original-ones/

https://www.space.com/spacex-starlink-satellite-deorbit-video

... while launching many new ones.

=46or example, there will be a =46alcon launch from Vandenberg, of severa= l dozen satellites, at 10:30 (or :40) PM Pacific time on Wednesday night = (though there are conflicting reports, and launches often get canceled):<= br />
https://www.edhat.com/news/spacex-starlink-launch-rescheduled-for-tuesday=

https://www.spacelaunchschedule.com/category/vandenberg-sfb/

https://www.space.com/spacex-starlink-group-3-4-launch-rocket-landing

https://www.ksby.com/news/local-news/spacex-targets-tuesday-night-for-fal= con-9-launch-of-starlink-satellites

A late evening launch time makes for good viewing because it's dark enoug= h to see the launch from a distance, and the rocket hits sunlight at the = edge of space, where exhaust moves outward in all directions uncontained = by atmosphere, leaving a tubular trail in the sky.

Here is a collection of screen grabs from a camcorder recording of a laun= ch in 2005: https://www.flickr.com/photos/docsearls/albums/999576 This la= unch will be later in the evening, but still quite visible. One big diffe= rence will be the return trip of the first stage to a platform out in the= ocean. I caught one of those in this series of shots here: https://www.f= lickr.com/photos/docsearls/albums/72157701027229232

(=46orgive my indulgence in space-freakery. I do enjoy this stuff, and I'= m not here in Santa Barbara often enough. But I am here now, so I'll be s= hooting it again, this time with a new camera and a longer lens.)

Doc

On Aug 30, 2022, at 9:53 AM, David P. Reed = via Starlink <starlink=40lists.bufferbloat.net> wrote:

I have no clue why this matters (other than this is in color).

The phased array antennas used by Starlink are quite limited - in particu= lar, there are 4 on each satellite and each earth-ground path is half-dup= lex, TDM, essentially. Limited by hardware. The problem of signal equaliz= ation and quantization limits prevent =22space division multiplexing=22 a= nd =22frequency division multiplexing=22 in practice.

The 4 msec =22turnaround time=22 at the physical level (satellite) means = that time from a packet arriving at one end to be sent to the other end o= f the sat-dishy links gets worse the more dishys are served by one of the= 4 antennas on the satellite.

trying to increase the coverage of an individual satellite basically mean= s serving more dishys per satellite, with less total bit rate, and much l= onger latency due to the half duplexness.

Now if the total bit rate of a sat-to-dishy link were, say, 1 Gigabit, li= ke an 802.11ac AP gives you, and the turnaround time were under 1 microse= cond rather than 4 msec. maybe then you could get reasonable Internet ser= vice to dishys.

But 240 Mb/s or 172 Mb/s as proposed for getting a bit more coverage per = satellite=3F This is nowhere near competitive with what we expect in the = US.

Sorry to rain on all the techy dreaming.

=46irst, it's worth looking at all the problems currently in Wi=46i perfo= rmance when you share an AP with multiple active stations using 100's of = Gb/s on the average (not just occasionally).

Dave - you tried in =22make-wifi-fast=22, and the architecture gets in th= e way there. (yeah you can get point to point gigabit/sec single file tra= nsfers, but to do that you invoke features that destroy latency and intro= duce huge variability if you share the AP at all, for these reasons).

Starlink is a good =22last resort=22 service as constituted. But fiber an= d last few-hundred meters wireless is SO much better able to deliver good= Internet service scalably.
Even that assumes fixing the bufferbloat that the Starlink folks don't se= em to be able to address...
=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F= =5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F
Starlink mailing list
Starlink=40lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/starlink

=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F= =5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F=5F
Starlink mailing list
Starlink=40lists.bufferbloat.net
https://lists.bufferbloat.net/listinfo/starlink
--630e6e7e_26f324ba_2eb--