[LibreQoS] Rain Fade (was Ack-filtering)

Wed Oct 26 09:27:56 EDT 2022

Tarana makes me chuckle a bit. Every few years, various WISP forums fill up
with "game changer, must buy X!" and it's tulip mania
<https://en.wikipedia.org/wiki/Tulip_mania> all over again. Tarana is
really expensive, and with the typical 3-5 year life cycle of wireless
products it's pretty hard to justify the cost if you have any sort of
competition, or lower population densities. Even if it weren't buggy, it's
a tough one. They also have a relatively limited window, because their
special sauce is pretty amazing but it's not *that* far ahead of what you
can do with 802.11AX and the sync extensions (and the Quantenna chips they
use just got discontinued, which is going to cause them - and Mimosa - a
lot of pain). And like all excited tulip purchases, we're already starting
to see people complain who bought it for areas that aren't the type of
deployment for which Tarana really shines.

Terragraph is nice on paper, but it's really over-engineered. (Not that
surprising if you've ever looked at Meta's "Folly" and other code; Meta
doesn't do "keep it simple") It does look great for getting coverage out
over a high-density area.

And that's the big problem with the current wave of tech. Ranges are
getting shorter, because competing with fiber requires MUCH higher CINR
numbers. That's *great* for the high-density areas (which tend to be
getting fiber anyway), but it's problematic for the really rural
deployments. It's really common out here to have to hop 10-15 miles between
clusters of buildings, and you're still only hitting 10 houses within a
couple of miles of a POP. So "decent service at longer ranges" is a lot
more useful here than "amazing service at 1.5 miles". That's also where
they are crying out for service (we've had so many customers "leave" for
StarLink and be back within a month when they find out what high-latency,
frequent disconnect 1d6+1 * 10 Mbps service feels like) - and the fiber
companies don't want to go. We've been working with an electric co-op that
rolled out fiber to provide wireless for the areas they don't plan on ever
building into (for example, 3 houses that connect to a fiber-served road...
with a 5 mile driveway).

On Tue, Oct 25, 2022 at 10:26 PM Dave Taht <dave.taht at gmail.com> wrote:

> On Tue, Oct 25, 2022 at 7:22 PM dan <dandenson at gmail.com> wrote:
> >
> > Tarana is a mess.  On one hand there's magic in their multipathing and
> reassembly, frankly the next best thing is 10% as good.  On the flip side,
> it's a highly manipulated transport and virtually none of your typical
> tools for monitoring connects works all that well because there's so much
> prioritization and filtering going on.  Next wisp up the road is demoing it
> out and talking about it a lot.    LOTS of bugs.
>
> Thank you as always for the torrent of information. I've been trying
> to get them to give fq_codel and cake a shot, they
> haven't listened.
>
> >
> > Terragraph is just a transparent fabric.  Like doing a big RSTP mesh of
> switches.  How this comes back to shaping is that it's necessary to shape
> for that first hop out since the second and third etc hops are all
> essentially the same speed.  Have to produce an artificial bottleneck at
> the head end PER DN to keep that in shape.  Also the complication of having
> 1.8-3.6Gbps 'backhauls' or worse, 3.6-7.2Gbps with channel bonding and dual
> radios on the POP DN.  Also, since you can have multiple DNs off your
> headend, each with that kind of bandwidth, now have to consider running a
> dedicated shaper box/instance able to handle that for each radio.
>
> This strikes me as a good goal for something like libreqos on the
> march to 100Gbit, or dedicated hw like what paraqum
> provides. I am dying to see if 100Gbit is feasible with 64 cores, now.
>
> > We're already selling 200,600, 1G services off these and they're
> incredible.  Smokes a docsis plant, we're 4ms to google on the headend and
> customers on an Eero Pro 6E will see 5-8ms on their devices.
>
> Docsis 4.0 LL has a much faster request/grant cycle available as a
> priority queue via the L4S concept. About 2ms. They want to sell low
> latency to gamers, and open up the diffserv bits so that they can bill
> both services trying to use that channel, and the customer.
>
> > As far as FSO, interesting but 80Ghz is basically better in every way
> and those radios are coming down in price pretty quickly.   10Gbps FDX link
> on siklu 8010 is under $6k if you have a good rep.  2.5Gbps <$3500.
> >
> > On Tue, Oct 25, 2022 at 10:57 AM Dave Taht <dave.taht at gmail.com> wrote:
> >>
> >> On Mon, Oct 24, 2022 at 4:25 PM dan <dandenson at gmail.com> wrote:
> >> >
> >> >
> >> >
> >> >>
> >> >>
> >> >>
> >> >> How bad are y'all's gear doing with rain fade on various techs and
> >> >> bands? in 08, in nica, I'd go from a working 70 db 10 mile shot to
> >> >> nothin at 5ghz when it rained, and I just laughed at the people
> trying
> >> >> to deploy 60ghz - but times change. I see a vendor trying to ship 60
> >> >> with *really good antennas* into the office market...
> >> >>
> >> >> big question to ask when so busy, please ignore me.
> >> >>
> >> >
> >> > I have extensive testing with almost every gear out there.
> >> >
> >> > 5Ghz, no appreciable fade in snow or rain.  Longest shot on network
> right now is 26 miles on AF5xHD 5Ghz on 2' dishes and we push a solid
> 300Mbps across this with zero fade.  Actually gets a tiny bit better in the
> rain, ie it is technically fading a bit but so is all the noise so it's a
> minor improvement.
> >> >
> >> > I have 2x 7 miles force 425 links that are pushing 550Mbps.  And a 10
> miles force 400c on 2' ubiquiti dishes that pushes 940 unidirectional in
> 80Mhz.  No rain fade.  Lots of af5xhd and force4x links in different
> distances.  We even mix in some LTU PtMP as PTP for price, ie LTU AP <>
> LTU-LR or LTU-Pro for PTP.  Works well enough though this product is
> susceptible to noise more than any other we use.
> >> >
> >> > Cambium 60Mhz cnwave is fantastic, legit 120 meters per link node to
> node or small CPE, 500M to big CPE, about 300 to the not-quite-released mid
> CPE.  Pushing 1.7Gbps FDX on against my preseem box and my m2 macbook with
> nperf UDP.
> >> >
> >> > Ubiquiti gigabeam line, <1km ok, <800m even better.  AF 'LR' and 'XR'
> rock solid at 2km, up to about 5km until they're down too much to be
> usable.  Always backed up by a 5Ghz radio.
> >> >
> >> > Ubiquiti Wave, legit AP<>CPE out 2km and never fails over.  4km w/
> wifi6 failover.  Fantastic product... probably the one to beat.
> >> >
> >> > Mikrotik 60Ghz 'ay about 200m on AP to small CPE, 500m AP to nRay.
> Can get a little more but it's really close and rain fade gets you.  These
> have 'ac wireless backup in them so we can EASILY push 300m on the small
> and 800m on the nRay knowing we have about 4 hours a year in 5Ghz failover.
> >> >
> >> > Basically, and MIMO 5Ghz, 6Ghz, or 2.4Ghz product isn't going to
> noticably fade.  MOST fade in these bands is actually thermal ducting
> 'turning' the beam off aim.
> >> >
> >> > 60Ghz should be considered 2 separate bands.  channels 1-4 are short
> range, <1km in PTP, <300m in PtMP if you want to have links stay up.
> channels 5,6 are 2-3x longer.  Unfortunately, only ubiquiti really playing
> in this space right now, mikrotik's channel 5 support is at a lower output
> power so it's 'ok'.  Tachyon coming into this space as well, but unproven
> and AFAIK zero beta deployments.
> >> >
> >> > 5Ghz <=200M service plans today with a well built network and good
> AP/Antenna choices, <=500Mbps with WiFi6 tech.  technically a bit more, but
> with reasonable ratios this is about right
> >> > 6Ghz <=900Mbps plans on live beta users.  OFDMA+MUMIMO is really
> delivering here.
> >> > 60Ghz 'low' band cambium, 1.7Gbps legit across the mesh, twice that
> with upcoming channel bonding.  base CPE 1Gbps port, mid 2.5gbps port, long
> 10Gbps port(s).  Build out model here is for 'In the rain' so no effective
> fade if built right.  if built wrong, fade to death.
> >> > 60Ghz 'high' band ubiquiti wave.  <=800Mbps.  Technically a bit more
> but I haven't convinced a Wave AP to a Wave LR to do it.  I can however get
> 2 customers/radios up to 1.5Gbps across the AP.  Plan with built in fade
> and intentional fail to 5Ghz beyond 2km.   Acceptable in Montana.
> >>
> >> Thx for such a deep dive! How about Tarana?
> >>
> >> How does "Terragraph" derived tech tie back into shipping products?
> >>
> >> I've been involved with a bunch of other highly speculative
> >> technologies over the years.
> >>
> >> # Free space optics
> >>
> >> Koruza showed promise at one point: http://www.koruza.net/
> >>
> >> Google's been trying to line up folk for this for a while:
> >>
> >> https://x.company/projects/taara/
> >>
> >> # Radios
> >>
> >> ... I appreciate the data on all these radios, their relative
> >> performance and range. This stuff needs to  end up in
> >> a wiki chart somewhere. ?
> >> ... but it's not knowing the buffering on em that bugs me the most. :/
> >> Got all kinds of measurement tools for that
> >>
> >> crusader, in particular, is coming along.
> >> packet caps from iperf sessions...
> >> flent to a raspi on the other side of the link...
> >>
> >> # Still a hardware guy at heart
> >>
> >> Back in 2013 I had thought hard about entering the market for radios
> >> in the 11ghz (and wispa just freed up a lot of space in the 10? or 12?
> >> band for future use) licensed spectrums, but I couldn't stomach
> >> working with qualcomm again. I have some hope a mediatek might get
> >> into that, and I could work with them.
> >>
> >> anyone got devices that can use channel 172 yet? That just got freed up
> too.
> >>
> >> >
> >> >
> >> >
> >>
> >>
> >>
> >> --
> >> This song goes out to all the folk that thought Stadia would work:
> >>
> https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-6981366665607352320-FXtz
> >> Dave Täht CEO, TekLibre, LLC
>
>
>
> --
> This song goes out to all the folk that thought Stadia would work:
>
> https://www.linkedin.com/posts/dtaht_the-mushroom-song-activity-6981366665607352320-FXtz
> Dave Täht CEO, TekLibre, LLC
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/libreqos/attachments/20221026/1393ac4c/attachment.html>