Comment by walrus01 | Hacker News Reader

Comment by walrus01 | Hacker News Reader

Parent Full thread

If you're referring to my comments, I'm probably the only person in this thread who works professionally in two-way satellite. My day to day work encompasses the venn diagram overlap between terrestrial ISP backbone network engineering, point-to-point high capacity microwave and millimeter wave band transport links, and various two-way satellite access technologies when a link via geostationary is the only viable option.

And I don't dislike Starlink. I actually think it's amazing tech. I'm criticizing the specific PR stunt of putting a starlink CPE at an exising ILEC CO, not what Starlink is.

Starlink, when ready for ordinary end user production use, will be an amazing improvement over cheap ($110-200/mo) consumer grade VSAT terminals. The physics and economics of launching 4000-6200 kg geostationary telecom satellites and their extremely limited aggregate capacity means that a consumer grade VSAT must be oversubscribed at a 32:1 ratio or worse. Resulting in terrible things like 60GB/month data quotas for end users. It's overall a miserable experience. Starlink has the potential to entirely disrupt and kill what I call the 'shitty VSAT' market.

Above the 'shitty VSAT' market there's a whole category of serious two way satellite technology that the ordinary consumer will never see. And things like o3b and starlink also have the potential to improve that. You would visibly cringe if you saw the monthly $ per MHz transponder costs for Ku spot beam capacity over eastern Africa. The dollars per Mbps ratio is ridiculously bad. In order to achieve any reasonable bps/Hz modulation density you can easily spend $100k to erect a pair of 4.5m size earth stations for a dedicated 1:1, SCPC link. And that's before you're looking at anywhere from $5000 to $30,000/month paid for transponder capacity, depending on how much data you're trying to move, and how asymmetric it is.

There's a reason why every telecom/ISP in every developing nation is trying their hardest to connect to the nearest submarine cable landing station and not remain reliant upon monthly C or Ku band transponder leases.

> There's a reason why every telecom/ISP in every developing nation is trying their hardest to connect to the nearest submarine cable landing station and not remain reliant upon monthly C or Ku band transponder leases.

It's not just cost, it's also the latency. While it's apparent you know this, most commenters might not realize the game changing difference with Starlink.

Starlink latency can be as low as 15ms to the nearest POP. The minimum latency for any geostationary satellite to the nearest pop is 528ms RTT. That's speed of light, going 28k miles 4x, assuming you and the POP are on the equator directly below the satellite.

yes, that as well. FYI minimum latency for a round trip ICMP ping on geostationary is more like 492-495ms. 520-525ms is also typical. Will vary somewhat with modem FEC complexity, FEC code rate, and exactly what modems are in use on a 1:1 dedicated link.

Amazing how just a nugget of bad info can change things.

Back in Satellite Controller school in 2003 we were taught geostationary is 28,500 miles. So even since then I've been using that as my reference point.

In reality, it's about 35,786km, or 22,236 miles[0].

That means the minimum RTT time is 479ms, given the perfect scenario I previously laid out and zero equipment delay. Which isn't a thing.

Then again, the mythical perfectly spherical radiation pattern dipole antenna that we base all dBi off of isn't a thing either.

0: https://en.wikipedia.org/wiki/Geostationary_orbit

One of the tradeoffs in modern state of the art VSAT modems is that you can intentionally choose more computationally 'expensive' FEC, to slightly improve your bps/Hz in a given fixed kHz/MHz amount of FDD transponder capacity, at the expense of higher latency. With the Comtech CDM-760 for instance.

This is getting better as the years go on and modem manufacturers are able to throw additional CPU and FPGA resources at FEC processing, without increasing their cost to build a modem a great deal.

35,786km above the equator.

Equator is 6,378km from the centre of the earth, so GEO is 42,164km, or 26,200 miles, so not sure where 28,500 came from.

> Then again, the mythical perfectly spherical radiation pattern dipole antenna that we base all dBi off of isn't a thing either.

It got knocked over by a spherical cow.

(I did have a colleague lose a satellite connection in Libya once which he blamed on a herd of cows. I think he aiming fairly low on the horizon to the east, with the disk near ground level, so I give him the benefit of the doubt)

One of mine is on 531ms at the moment, but that's guaranteed traffic on a dedicated slot. General wifi stuff on BGANs in background mode tends to be twice that.

By comparison, my office in Sydney is currently 276ms rtt from London.

"Starlink, when ready for ordinary end user production use, will be an amazing improvement over cheap ($110-200/mo) consumer grade VSAT terminals. The physics and economics of launching 4000-6200 kg geostationary telecom satellites and their extremely limited aggregate capacity means that a consumer grade VSAT must be oversubscribed at a 32:1 ratio or worse. Resulting in terrible things like 60GB/month data quotas for end users. It's overall a miserable experience. Starlink has the potential to entirely disrupt and kill what I call the 'shitty VSAT' market."

Actually, GEO satellites have quite a bit of aggregate capacity (1 Tbps and going up) and more capacity where it matters (over land, and specifically, population centers) than LEO constellations, and therefore have better bandwidth economics. They also last ~3x as long, which brings the $/Mbps down more.

60 GB/month caps are not really a thing anymore, unless you're still using HughesNet.

Latency will be dealt with using hybrid modems that can use LEO/DSL/4G for latency-sensitive applications and GEO for large amounts of cheap bits. Hughes, to their credit, has had such a solution available since 2007, last I checked.

> Actually, GEO satellites have quite a bit of aggregate capacity (1 Tbps and going up) and more capacity where it matters (over land, and specifically, population centers) than LEO constellations, and therefore have better bandwidth economics. They also last ~3x as long, which brings the $/Mbps down more.

the aggregate capacity of a 180 million dollar (construction plus launch cost) 6000 kg geostationary telecom satellite is less than two strands of fiber and a set of modern 100/200/400GbE coherent DWDM system line terminals.

saying that the throughput of all of the ka-band spot beams on a satellite is 1Tbps is both double-counting math for the capacity in both directions, and does not reflect the actual capability of the satellite. maybe if every single ground station was something large enough and with enough link budget to operate at 16APSK and better modulations. not a realistic figure if you're talking about low-cost, sub $1500 hardware, consumer grade ku and ka band vsats.

for a randomly chosen rural zip code in WA state, hughesnet is offering a 50GB quota/month for $150/month. so yes, 60GB transfer limits and things are still a real thing. and during peak evening hours TDMA contention means that latency will be 1150ms+, or worse.

the unavoidable economics of consumer grade small VSAT require serious oversubscription. Dedicated 1:1 satellite capacity starts from $1500/Mbps and up, and requires a serious VSAT terminal, not a hughesnet/viasat/wildblue type toy.

as to your last paragraph, if you have a proper 4G/LTE network in your area, just use that. There's verizon and at&t resellers with plans available that have quotas close to 200GB/month for a reasonable price now. why you would want a hybrid cheap vsat and 4G LTE setup, as a consumer, I don't know.

> the aggregate capacity of a 180 million dollar (construction plus launch cost) 6000 kg geostationary telecom satellite is less than two strands of fiber and a set of modern 100/200/400GbE coherent DWDM system line terminals.

Not sure how that's relevant. In any case, the total cost of a GEO is closer to double that.

> saying that the throughput of all of the ka-band spot beams on a satellite is 1Tbps is both double-counting math for the capacity in both directions, and does not reflect the actual capability of the satellite.

This is not correct.

> maybe if every single ground station was something large enough and with enough link budget to operate at 16APSK and better modulations. not a realistic figure if you're talking about low-cost, sub $1500 hardware, consumer grade ku and ka band vsats.

I'm not sure if you're talking about the ground stations that connect to fiber backhaul, here, or to the CPE terminals. In any case, ground station capacity is not an issue, and neither is CPE terminal capacity.

> for a randomly chosen rural zip code in WA state, hughesnet is offering a 50GB quota/month for $150/month. so yes, 60GB transfer limits and things are still a real thing. and during peak evening hours TDMA contention means that latency will be 1150ms+, or worse.

Yup, that's what I said: "... unless you're still using HughesNet."

> the unavoidable economics of consumer grade small VSAT require serious oversubscription. Dedicated 1:1 satellite capacity starts from $1500/Mbps and up, and requires a serious VSAT terminal, not a hughesnet/viasat/wildblue type toy.

Again, not sure how this is relevant. Every residential (and even many business) connectivity option is oversubscribed.

If you're saying that satellite cannot compete with fiber or cable, then you'd get no argument from me. If you're saying that satellite is a shitty option, well, I think you'd find that some people disagree with you on that.

That's not to say there isn't room for improvement, but providers know this and are working on it.

> In any case, ground station capacity is not an issue, and neither is CPE terminal capacity.

because consumer grade CPE terminals need to be so small (because of hardware cost and economics), link budgets suck and less complex modulations must be used. As compared to serious 1.8 meter and larger VSATs. Show me a consumer grade VSAT that's using 16APSK/32APSK and the correspondingly higher bps/Hz efficiency that can be achieved with big, expensive satellite earth station gear used for dedicated 1:1 links.

the aggregate throughput of a geostationary satellite is greatly reduced if you start using tons of tiny VSATs on it. Go look at how much actual capacity is available in a 15 MHz chunk of Ku transponder capacity (even with a 51dBm EIRP hot spot beam), used with an iDirect evolution series linecard at the hub end, that is populated by a bunch of cheap 1.2 meter compression molded fiberglass dishes with 3W/4W class BUCs on them. How many Mbps is that? Not a lot.

> Every residential (and even many business) connectivity option is oversubscribed.

the aggregate throughput capacity of a single CMTS port on a DOCSIS3 or DOCSIS3.1 network segment is equivalent to a significant portion of the entire capacity of a geostationary satellite serving consumers. one CMTS port can serve hundreds of (yes, oversubscribed) 200 to 1000 Mbps class residential end users. One 10GbE circuit feeding a 2U rackmount sized GPON OLT can serve a ridiculous number of end user customers. It's oversubscribed but not in such a way that individual users see latency increases during peak hours (unless somebody has really fucked up their network). With the CMTS example we're talking about things that regularly achieve 1024QAM to 4096QAM modulation on copper wireline plant using the RF channels that previously would have been analog TV.

wireline is oversubscribed? yes. I could show you traffic charts from a 175-suite condominium building where every unit is wired with symmetric gigabit 1000BaseT to the home, that has an 80% take rate of subscribers. The whole building has a single 10GbE uplink and everyone is perfectly happy. Because of course not everyone tries to use the whole capacity of anything all the time. But when you start getting into the wildly high oversubscription ratios needed to make geostationary consumer grade VSAT stuff work economically, it becomes a very poor end user experience.

> If you're saying that satellite is a shitty option, well, I think you'd find that some people disagree with you on that.

I'm saying geostationary consumer grade satellite stuff is a shitty option and should always be the choice of last resort when anything else exists, due to its unavoidable poor economics and aggregate capacity.

I have a lot more optimism that starlink will have the real world throughput needed to be a viable option for rural broadband. Because the link budgets are much better with small antennas and small tx poweer electronics, due to the LEO orbit. And the much lower latency. I think people who believe starlink will replace their 800 Mbps comcast DOCSIS3 32x8 connection or GPON last mile gigabit connection are not fully informed. Properly implemented wireline networks will always have a great deal more throughput at a much lower deployment cost. What Starlink will kill is crappy consumer grade VSATs and sub-75Mbps class terrestrial PtMP WISPs.

> Show me a consumer grade VSAT that's using 16APSK/32APSK...

Not that it matters, but, I could.

> How many Mbps is that? Not a lot.

So, e.g., Viasat offers 100 Mbps service.

> the aggregate throughput capacity of a single CMTS port on a DOCSIS3 or DOCSIS3.1...

Yes, you can get tons of traffic through a coax cable. There's more bandwidth and you can use higher efficiency modulations because the SINR is way better. Not sure how this is relevant, though.

I'll point out, though, that plenty of people see slowdowns on their cable modem networks during peak busy times. Even these networks have their limits.

> ... it becomes a very poor end user experience.

Again, I think some people would disagree with you on that.

> ... real world throughput needed to be a viable option for rural broadband. Because the link budgets are much better with small antennas and small tx poweer electronics, due to the LEO orbit. And the much lower latency.

No doubt about the lower power electronics (vis-a-vis GEO) and the lower latency, but even so, SL's CPE is around $1,000, and the reliability through rain, wind, and snow is unknown. Setting aside the reliability question for now, I think what it really comes down to is the price vs latency trade-off. In order to be profitable, SL will have to be more expensive than traditional GEO solutions because the CPE, the ground network, and the actual constellation all cost more on a $/bit basis than GEO solutions. It will be interesting to see if the lower latency will be enough to draw consumers.

The big X-Factor in all of this is how much more capital Elon can raise. He has an amazing talent for raising money and if retail investors don't care that SL is less cost-effective than other options, then it can go on for a long time.

Reply on news.ycombinator.com