Hacker News new | past | comments | ask | show | jobs | submit login

> SpaceX is quietly working toward nothing less than a revolution in global data infrastructure

I mean, that's a really cool 40,000 ft view idea and all, but global data infrastructure is made up of things like the Hibernia Atlantic cable, its several dozen cousins of post-2000 transatlantic and transpacific cables with DWDM terminals on both ends, and cool things like 100, 200, 400Gbps per wavelength coherent QPSK/16QAM DWDM terminals. And major terrestrial traffic exchange points of existing Internet infrastructure where ISPs put $150,000 core routers (example: 60 Hudson, NAP of the Americas, Telehouse Docklands, Otemachi Building in Tokyo, One Wilshire, 350 E. Cermak, etc).

Satellite traffic is a pretty tiny drop in the bucket compared to terrestrial backbone infrastructure. It's a truly admirable goal to bring affordable broadband to really impossible to reach locations via satellite. On the other hand there are a lot of emerging terrestrial WISP technologies and PTP microwave technologies that can be used in rural areas to provide bandwidth without the need to pipe it up into space and back. It's a lot cheaper to establish a tower site on top of a mountain, even if you need to bulldoze a road to the top, and spend $30,000 on routers and PTP/PTMP radio gear.

In some of the areas of the rural western US where my network engineering job touches, there are WISPs which are rapidly eating into the customer base of people who are (rightfully so) dissatisfied with highly oversubscribed consumer grade VSAT satellite systems.

In the end there will be a combination of many things. The next generation of high capacity Ka-band geostationary satellites (ViaSat-2, etc) are a lot higher capacity. Services like o3b allow ISPs to buy a dedicated 1:1 high capacity pipe to places that can't be reached by PTP microwave and are uneconomical to reach by submarine or terrestrial fiber (example, all of o3b's new pacific island nation state customers). There's traditional geostationary c and ku band capacity from SES, Eutelsat, Intelsat, AsiaSat, russian companies, etc. And of course terrestrial fiber. You don't need a huge amount of money to run singlemode these days, assuming aerial wood poles and a mostly rural area, you need two guys, a bucket truck and about $10,000 worth of tools.




Completely agree on WISPs. I think SpaceX will be pursuing mesh networking too.

Could an OTA update (and repointing 90°) turn a satellite terminal into an auto/calibrating point-to-point MIMO link?

>global data infrastructure is made up of things like the Hibernia Atlantic cable

Surprisingly the stated primary goal of the SpaceX constellation is actually to compete with long-distance fiber backhauls. Their sell for satellite backhaul is that it's lower latency (no need to avoid continents, 50% faster speed of light, fewer hops) and works everywhere.

Giving global gigabit internet to rural areas and ~10% of urban customers (with the rest on fiber) is only a bonus. :)

Most of what we know about the plan comes from this video. Timestamp is to the start of the juicy bits. https://www.youtube.com/watch?v=AHeZHyOnsm4&t=2m10s


IMHO the shannon limit and basic laws of RF/path loss and channel capacity say that satellite is not a true competitor to backbone links by singlemode fiber. For the equivalent of "last mile" services, yes, but not as a replacement for laying fiber between points A and B.

The entire data throughput capacity of a current generation, 5500 kilogram, geostationary Ka-band satellite that costs $185 million to build and launch is much, much less than the 80 channel x 100 Gbps per channel DWDM system you can run on two strands of 9/125 singlemode fiber. And vastly less than the 144, 288 or 864 strand count fiber cable you would see laid between two cities by a carrier-of-carriers operator like Zayo these days.

It is fabulous to see more competition for high priced monthly-leased transport kHz/MHz from geostationary satellite operators. O3b was an amazing thing (and still is). More competition is good. But it's a pipe dream to say that satellite backhaul will ever be preferable to fiber carrying N x 10GbE circuits or a 100GbE circuit...

https://en.wikipedia.org/wiki/Shannon%E2%80%93Hartley_theore...

https://en.wikipedia.org/wiki/Noisy-channel_coding_theorem


I really appreciate your comments. This stuff needs to be out there more.

I expect crosslinks and backhaul up/down will be multipath laser w adaptive optics, not RF. As you say the physics demands it.

At 1100 km altitude each satellite has vacuum line-of-site to any other satellite within 6000 km (ground track). By "skipping" satellites you gain extra bandwidth capacity and reduce latency. Easy to route around damage, with no single point of failure (unlike fiber in certain areas). Obviously this will all be optimized with network and timing analysis to hell and back, just like fiber.

LEO has advantages of lower distance traveled, dramatically lower attenuation, faster speed of light, fewer hops, no cable breaks to fix, and no actual cable to run (which the expensive part of fiber, after all). You just build the repeaters, and exploit the fact that the exosphere is really transparent.

Musk, a guy who knows his physics and math, predicts in that youtube video that they'll ultimately do "more than half" of all long distance traffic. He also acknowledges that they have to 'skate to where the puck will be' re: telecommunication technology or they'll end up dead like innumerable predecessors. It's an interesting watch.


> IMHO the shannon limit and basic laws of RF/path loss and channel capacity say that satellite is not a true competitor to backbone links by singlemode fiber. For the equivalent of "last mile" services, yes, but not as a replacement for laying fiber between points A and B.

The filing states they'll be using free space optics / lasers between satellites. The Ka/Ku links are only for the initial uplink and downlink.

> But it's a pipe dream to say that satellite backhaul will ever be preferable to fiber carrying N x 10GbE circuits or a 100GbE circuit...

It is not a pipe dream. Free space optics in... well space, have a 50% propagation latency advantage vs terrestrial fiber. This helps equalize things somewhat.


> The Ka/Ku links are only for the initial uplink and downlink.

Then there's a huge bottleneck, if the links from the satellite constellation as a whole to the trunk earth stations (not the CPEs) are high capacity Ka-band, there's RF issues with capacity...

It's like if you have a network that's composed of a whole lot of 10GbE backbone links from router to router and your IP transit connection to upstream ISPs/the global v4/v6 routing table goes through one 1000BaseLX link.


No, it's nothing like that. They're talking about highly localized signals via phased arrays on both sides. Additionally you underestimate the capacity of wireless: LTE has no problem doing 30 bits per 1 hz of spectrum. These will run at a lot more than a 1 gbit globally shared last mile.


Keep in mind that you don't really need to compete with fiber if you believe there are enough unserved people. I think his comments were misleading since they probably won't target that market. Another problem with Leo is that the rate of consumption is going up extremely fast on the internet. By the time this launches the total capacity of their system may sound like a lot, but it's spread over all the satellites uniformly. So what do you have to do if you want to double your capacity? You need to launch another 4400. They're essentially going to play a game where they need to keep up launches fairly quickly to go with the rate of internet consumption. I also believe the ground cost will be interesting, especially the user terminal. Typically phased array is so expensive that they'll need to sell it at a loss.


Can it really be faster than fiber? The majority of traffic flows somewhere between Europe, North America, and East Asia. The only part where submarine cables take a detour is Europe<->Asia. US<->Europe and US <-> East Asia are fairly direct. Why is the speed of light faster for satellites than fiber? Even if the satellites are "just" a few hundred kilometers above earth, I can't see how they can have a shorter (and thus faster) way.

That aside, I don't think there's a big market for even lower latency (apart from algo traders). It's bandwidth that matters. You can get stable <150ms round trip times Europe<->East Asia and much shorter times to the US already. Problems with servers on other continents are not due to high latency but low throughput. And as others have pointed out, I don't think a system of satellites can compete with enormous bandwidth a single sea cable can provide.

Does anyone know if there are actual numbers out that show how satellites could transfer even a fraction of what's already travelling below surface?

Rural areas will definitely profit. But while the goal is great (internet for all), it's probably not what pays the bill. Aren't FB/Google's ideas of planes/balloons cheaper?


> Why is the speed of light faster for satellites than fiber?

The speed of light in a medium is slower than the speed of light in vacuum. Fiber commonly propagates at just 200e6 km/sec.

Latency is still important because it's a significant limitation of GEO satcom.

> Does anyone know if there are actual numbers out that show how satellites could transfer even a fraction of what's already travelling below surface?

I'm unaware of any fundamental physical limit that prevent optics in space from matching or exceeding the bandwidth of terrestrial fiber. In terms of engineering the main limit is likely keeping beams in precise alignment.


Thanks for the info with the fibre, thought it was closer to light speed. But for a satellite the signal would also travel through the atmosphere for most of the time, or am I mistaken? Is the speed through air close to the vacuum speed? Otherwise it's probably just 10-20% which I guess would be slower after considering the extra distance?


>Is the speed through air close to the vacuum speed?

Yes, very close. The index of refraction of air is 1.0003, so the speed of light in air is c/1.0003 = 99.97% the speed of light in a vacuum.

The index of refraction in a doped silicon telecommunications fiber core is around 1.4475, so the speed of light is 1/1.4475 = 69.08% the speed of light in a vacuum.

>Otherwise it's probably just 10-20% which I guess would be slower after considering the extra distance?

Undersea fibers have to avoid these things called 'continents.' For long distance hops this makes satellite the fastest system that's physically possible. https://personalpages.manchester.ac.uk/staff/m.dodge/cyberge...


Thanks for the thorough explanation!

Regarding the sea cable length: the continent argument is what I was referring to before. I don't think it's valid as most data flows US<->Asia or US<->Europe. And in both cases the ways are nearly direct. Only Europe<->East Asia has a major detour but I don't know if that warrants a global satellite system. One could still put a cable through russia (actually wondering why that doesn't exist for algo traders, connecting HK and London directly).


WHOOPS, quite the typo there. It's 200e3 km/sec.


True, but submarine cable bandwidth is ridiculously expensive many places. E.g. I know locations where the only local cable operator charge $3000/month for 20Mbps uncontended bandwidth, just because they can. Meanwhile in London even the most extortionate colo-provider I work with charge "only" $250/month for similar quality bandwidth.


Compared to current costs of geostationary dedicated 1:1 capacity (SCPC), 20Mbps full duplex for $3000 is actually an amazingly good deal. If you're in a place in Africa with no fiber access and your only options for a high bandwidth trunk link to the world are c band satellite, you're looking at $25,000+ to build your own 3.0 meter+ sized earth station and then $1800 to $3000 per individual Mbps per month for leased transponder kHz to reach a teleport in France or Germany. This is one of the problems o3b was designed to solve.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: