Fibre versus Starlink Discussion

[ItsNotAboutTheMoney]

I am saying that. Sure with a static screen only editing code it's okay, but when you switch tabs or windows etc. a large amount of data needs to be sent. Clicking a link in the browser creates huge delays as the whole screen has to refresh.

You're using 4k, which the vast majority of people don't have or need.

In any case, as for load over the next few years, well the current generation of consoles are already coming in digital only versions with games already in the 60-70GB range for the initial install. We can expect that to keep growing.

Are you suggesting that we should subsidize rural broadband for gamers?

Besides 60GB ~= 60,000MB. If you can download at only 10MB it's less than 2 hours to download. Oh no, how is a gamer going to be able to wait that long! That's almost like the bad old days when you'd have to go into The Big City to buy stuff.

Note that we regularly drive for an hour each way to visit with friends or go eat and we actually live in a city. My state is over 1/3 the size of the UK, but with 1/40 the population and half of the population is in the south of the state, so density really drops for a large part of the state, and we're not even a large state. Montana, for example, is larger than the UK and has a smaller population than Maine.

As someone once said, Americans think 200 years is a long time and British people think 200 miles is a long way. I adjusted my expectations when I moved here. In the UK I didn't even own a car.

Do you have any experience with rural living?

YouTube 4k videos are clocking in around the 40-50Mb/sec range, which will continue to increase as 120 fps and 8k become more common. And then you have VR and streaming games. You may think these things are unimportant but they all go towards creating a gap, and as new technologies become available it will only continue to increase.

YouTube and other streaming services automatically adjust video quality to bandwidth.

And that will never stop, because customers are and will be streaming to mobile devices with difficult capabilities and connections.

Remember that Starlink will continue to get slower as more users are added to the system. It's shared bandwidth over the air, right now the low number of users is an advantage but that will slowly fade.

Sez you.

 

[ItsNotAboutTheMoney]

Then why is latency on Starlink so poor? On fibre I'm seeing sub 1ms pings to sites like Google, which is a small fraction of the time it takes the signal to even get to the satellite.

It should hopefully be obvious that if it has to go first up and then back down it's never going to be as fast as taking a more direct route to the nearest datacentre.

Of course it's not as fast. But you don't _need_ that latency to have essential access. My wife is doing Teams meetings every day on ADSL which never has latencies anywhere near as low as fast fibre.

There are two esssential needs for broadband services:
- information distribution
- communications

Neither needs either massive bandwidth or ultra-fast ping times.

Starlink's standard latency is like good ADSL, with slowdowns related to the currently limited constellation. As the constellation fills in, it will be more consistent.

If you can affordably reach a rural community, and get wide-scale buy-in, as B4RN has done, great, businesses with a reasonable model can go ahead and government can help underwrite it.

But as the densities continue to fall, the cost per customer rises, and the risks of connection loss increase. That's where wireless delivers better value. That's why developing nations have skipped landline, and it's why mobile telecommunications has destroyed the economics of landline telephone services, to the point that telecommunications companies with landlines are begging regulators to remove the requirement to serve rural customers.

 

[bkp_duke]

Then why is latency on Starlink so poor? On fibre I'm seeing sub 1ms pings to sites like Google, which is a small fraction of the time it takes the signal to even get to the satellite.

It should hopefully be obvious that if it has to go first up and then back down it's never going to be as fast as taking a more direct route to the nearest datacentre.

Because Google has distributed servers in many locations you would not think are "traditional" datacenters. I've even seen Google DNS servers in the bases of larger cell phone tower distributions. If you are getting < 1ms ping to google, then you are hitting one of those.

You picked a VERY edge case to make your argument (pun intended for those that know what edge computing is).

I live in San Diego, I run datacenter operations for 6 DCs in the US and one in Europe, here are my ping times (on fiber) to different locations:
Google.com (local server) - 6ms
San Francisco (our datacenter) - 17ms
Seattle (our datacenter) - 39ms
Dallas (our datacenter) - 41ms
Chicago (our datacenter) - 50ms
Washington DC (our datacenter) - 65ms
Amsterdam NL (our datacenter) - 153ms

The roundtrip latency for Starlink to your nearest groundcenter is currently about 30ms, with an aim of 20ms when the network is built out.

Simply put - that is comparable to fiber.

 

[bxr140]

...as for load over the next few years...

Skipping past the obvious disconnect between edge cases and average cases and the otherwise seemingly self-evident incremental (= not huge step functions) growth in load "over the next few years", let's try this from a different direction:

How much will it cost to run fiber to everyone in the US?

Remember that Starlink will continue to get slower as more users are added to the system.

Remember that Starlink can add capacity by adding satellites. That's why they've filed for a gazillion (+/-)--not because they need them right now or even that models say they will need them in the future, but because there's the possibility of needing some of them sometime in the future.

Also remember that the version of the starlink satellite going up now will be long gone by the end of the decade. |f there's one thing that's a pretty safe bet, SpaceX will leverage technology evolution into higher processing power and smaller beam widths, so even if they're not actually putting more power to the ground (which is regulated by the filings and a PITA to modify), they will effectively be putting more power to each user.

Also also remember that nobody is expecting Starlink to replace typical terrestrial service. As noted upthread it is a complement to terrestrial, providing service to those with crappy or no options. If one lives on typical cable service, one will be better served by that cable than Starlink...so there's a natural limitation to practical user base for satellite internet. As it were, the sky is not the limit.

Finally, also also also remember that as time goes on terrestrial services (including MNOs; witness the recently opened auction) will expand their reach and performance as well, which will actually decrease demand for Starlink in the most difficult to serve areas--areas that are not quite populated enough to currently justify incumbent infrastructure, but dense enough to run into beamsharing limitations. As time goes on, likely even accelerated by pandemic/WFH migrations, rural hamlets that didn't previously have enough customers/demand to justify investment from the incumbent may cross the line into profitability with extended/improved infrastructure, ostensibly relegating Starlink to the #2 option.

 

[hmcgregoraz]

Then why is latency on Starlink so poor? On fibre I'm seeing sub 1ms pings to sites like Google, which is a small fraction of the time it takes the signal to even get to the satellite.

It should hopefully be obvious that if it has to go first up and then back down it's never going to be as fast as taking a more direct route to the nearest datacentre.

As other have said, you are looking at a very distributed service (google), and not at long distance communications. I was clear that the latency benefits were for specific long distance links, where the ISL link performance will balance out with the up and down latency hit.

Here are several videos that explain the potential latency of startlink, both with ISL and without ISL:

-Harry

 

[scaesare]

Then why is latency on Starlink so poor? On fibre I'm seeing sub 1ms pings to sites like Google, which is a small fraction of the time it takes the signal to even get to the satellite.

It should hopefully be obvious that if it has to go first up and then back down it's never going to be as fast as taking a more direct route to the nearest datacentre.

"Poor latency" != >1ms RTT

 

[banned-66611]

"Poor latency" != >1ms RTT

Not today... But actually you are right for once, there is another component: Variability. Anything that uses RF will suffer from that.

That's why many games already have a "wifi filter" that blocks players on wifi or with excessively variable latency from ranked matches. Other players don't want to be disadvantaged by having to play against them.

 

[banned-66611]

Because Google has distributed servers in many locations you would not think are "traditional" datacenters. I've even seen Google DNS servers in the bases of larger cell phone tower distributions. If you are getting < 1ms ping to google, then you are hitting one of those.

The speed of light is way, way, way after than you think.

In a vacuum it's 300km/ms. It's around 200km/ms in fibre. Plus switching, however long it takes the sever to respond.

So you can get 1ms ping times to any server within about 75km of your location on fibre.

To get latency as poor as Starlink you would need to be over 1,200km from the server, probably over 1,600km.

Don't forget that Starlink is hampered not just by the distance (340km up best case with the satellite directly overhead) but also by the nature of it being a shared service with limited bandwidth split up in the time domain.

 

[bkp_duke]

The speed of light is way, way, way after than you think.

In a vacuum it's 300km/ms. It's around 200km/ms in fibre. Plus switching, however long it takes the sever to respond.

So you can get 1ms ping times to any server within about 75km of your location on fibre.

To get latency as poor as Starlink you would need to be over 1,200km from the server, probably over 1,600km.

Don't forget that Starlink is hampered not just by the distance (340km up best case with the satellite directly overhead) but also by the nature of it being a shared service with limited bandwidth split up in the time domain.

Sorry, but this is just not correct. I am in this industry, and just pinged servers that I know in San Diego (40 miles south of me).

I am on fiber, and a damn nice version of it too. All ping times to those servers were 4-7ms.

Simply put, you are making fiber out to be better than it is (and it's really good).

 

[scaesare]

Not today... But actually you are right for once, there is another component: Variability. Anything that uses RF will suffer from that.

That's why many games already have a "wifi filter" that blocks players on wifi or with excessively variable latency from ranked matches. Other players don't want to be disadvantaged by having to play against them.

OK you are a bit all over the place, so here goes:

Not today...

Yes, and designing products/services that fill a need today while holding the prospect of improving to meet future needs is not only viable, it's smart.

Incidentally, if sub-ms latency become a requirement tomorrow, fiber isn't an option for anything that's not in a colo down the street either.

But actually you are right for once

Firstly, chill on the insults (implied or otherwise)

Secondly, I've not chimed in on this topic previously, so please check your facts before firing shots, mmkay?

Thirdly, I'm glad you see that touting a latency spec that isn't realistically achievable for using real-world services even on fiber is not an argument against the feasibility of Starlink.

there is another component: Variability. Anything that uses RF will suffer from that.

(giving you the benefit of the doubt you are talking about the Starlink RF ground/terminal link, and not about actual local WiFi, which is completely out of scope for this discussion)

Here's what else induces variability: network routing changes, pipe congestion, switch load, router load, server response time due to load, etc...

No rea-world service has a static RTT. That's why there are acceptable thresholds, not absolute numbers. The 20-30ms being targeted is reasonable for a vast number of applications.

That's why many games already have a "wifi filter" that blocks players on wifi or with excessively variable latency from ranked matches. Other players don't want to be disadvantaged by having to play against them.

See above.

Also, gamers are not the only broadband customers.

I currently get 18-20ms response times over VPN to my work site. You know what I do over that link? Videoconference just fine.

 

[Matias]

The speed of light is way, way, way after than you think.

In a vacuum it's 300km/ms. It's around 200km/ms in fibre. Plus switching, however long it takes the sever to respond.

So you can get 1ms ping times to any server within about 75km of your location on fibre.

To get latency as poor as Starlink you would need to be over 1,200km from the server, probably over 1,600km.

Don't forget that Starlink is hampered not just by the distance (340km up best case with the satellite directly overhead) but also by the nature of it being a shared service with limited bandwidth split up in the time domain.

In longer distances Starlink could be faster than fibre if Starlink’s satellites have sat to sat laser links?

 

[bkp_duke]

In longer distances Starlink could be faster than fibre if Starlink’s satellites have sat to sat laser links?

Substantially, but for shorter distances fiber will remain king.

Watch the 3 videos posted yesterday in this thread, there are some latency calculations there. Latency will beat fiber as the distance gets longer, but for shorter distances, the advantage reverses.

 

[banned-66611]

Sorry, but this is just not correct. I am in this industry, and just pinged servers that I know in San Diego (40 miles south of me).

I am on fiber, and a damn nice version of it too. All ping times to those servers were 4-7ms.

Simply put, you are making fiber out to be better than it is (and it's really good).

What can I tell you, I can ping servers 50km away and get sub 1ms response times. Maybe your servers are taking a few milliseconds to respond? Maybe the network they are on has a complex firewall that adds some latency? Maybe the fibre takes a very round-about route?

How else would you explain it? The laws of physics somehow work differently in San Diego?

 

[banned-66611]

Substantially, but for shorter distances fiber will remain king.

Watch the 3 videos posted yesterday in this thread, there are some latency calculations there. Latency will beat fiber as the distance gets longer, but for shorter distances, the advantage reverses.

That assumes that they will route longer distance traffic over laser link. Even if they get the laser links working one day they will have limited bandwidth and most traffic will likely be just sent back down to the surface for routing.

Having said that these days it's less important as most stuff is cache on a CDN somewhere near you anyway.

 

[bkp_duke]

That assumes that they will route longer distance traffic over laser link. Even if they get the laser links working one day they will have limited bandwidth and most traffic will likely be just sent back down to the surface for routing.

No, it does not. Watch the videos. Even with ground routing long distances are still faster with Starling.

 

[bkp_duke]

What can I tell you, I can ping servers 50km away and get sub 1ms response times. Maybe your servers are taking a few milliseconds to respond? Maybe the network they are on has a complex firewall that adds some latency? Maybe the fibre takes a very round-about route?

How else would you explain it? The laws of physics somehow work differently in San Diego?

I explain it that you are not being truthful. Plain and simple.

 

[bxr140]

The satellite-to-satellite laser links might not be real. They haven't actually demonstrated them yet, and I don't believe that the current satellites have the capability.

Prototypes are real. Optical cross links in space were a thing before Spacex (primarily in defense applications, because it turns out its real hard to jam a laser). The real exercise for Spacex is to productionize terminals down to a cost basis that aligns with the Starlink mission goals. There is no evidence to suggest SpaceX will not get the OISL network functional for production service, the only question is when.

That assumes that they will route longer distance traffic over laser link. Even if they get the laser links working one day they will have limited bandwidth....

Laser links are all but a non-issue from a traffic volume perspective, not to mention they have the added benefit of being unregulated as opposed to, say, the Iridium RF cross link network (which requires ITU approval to operate at specific frequencies and power levels)...regulation which of course limits or at best complicates future expansion. I don't know what the last public figure is for a Starlink ISL, but you can go buy a Mynaric or Tesat unit that can do bidirectional 10gbps in LEO, give or take depending on range. Given that the ranging behind those rates is WAY higher than the XXXX satellite starlink constellation (they're based on XXX satellite constellations), it is a very plausible logic step to expect Starlink units to be able to pump a similar amount of traffic, if not more. If each starlink sat has 4 heads pumping 10gbps, that's ~double that sat's user link capacity and, when dropped as a node in a satellite constellation (say, over the Atlantic) one could imagine a number of pathways that sum up to a substantial amount of traffic.

...most traffic will likely be just sent back down to the surface for routing.

Yes of course, a good portion of a sat's traffic will just go up and down through the user/feeder links, because a good portion of the traffic volume will be regional (like streaming). That in no way reflects on the theoretical limit or likely realistic capability of the ISL network.

As others have established in this thread, the latency of that up and down trip, even when subsequently leveraging terrestrial paths, is more than adequate for low latency demands like video calls.

 

[BioSehnsucht]

The satellite-to-satellite laser links might not be real. They haven't actually demonstrated them yet, and I don't believe that the current satellites have the capability.

SpaceX's Starlink internet shows fast speeds during early tests, capable of gaming and streaming

Tice also said that the company recently conducted a test to see if it can connect two satellites in orbit with what SpaceX calls “space lasers.” Also known as intersatellite links, the space lasers would improve the Starlink network by allowing it to exchange data in between the satellites in orbit, rather than beaming it back-and-forth to the ground. SpaceX tested two of the Starlink satellites in orbit that are equipped with the intersatellite links.

“With these space lasers the Starlink satellites were able to transfer hundreds of gigabytes of data,” Tice said.

She added that, once SpaceX has “space lasers” working consistently throughout the network, “Starlink will be one of the fastest options available to transfer data around the world.”

It's currently only on two satellites equipped with prototype hardware, but it has been tested to work in space, so it's just a matter of ironing out any issues, scaling up hardware production, and installing them on future flights (and eventually replacing the existing satellites until they all have it, which may take a year or two, but likely not longer than that).

 

[BioSehnsucht]

What can I tell you, I can ping servers 50km away and get sub 1ms response times. Maybe your servers are taking a few milliseconds to respond? Maybe the network they are on has a complex firewall that adds some latency? Maybe the fibre takes a very round-about route?

How else would you explain it? The laws of physics somehow work differently in San Diego?

Most fiber networks are not following the shortest path, but weird paths along various rights of way (such as old rail lines, or other utilities). So other than last mile you can pretty much guarantee it'll never be shortest path. Plus, usually you'll pass through several switches / routers on the way to the destination, which may add variable latency as well, so most people will see >1ms latency to even nearby servers, with the exception of things like google search or CDN nodes that are often colocated much closer to the end user inside the ISP network, rather than being at a datacenter.

It's not impossible that you get the latency to the distance you state, but most people wouldn't. And so what if Starlink isn't a good match for you? That's fine. The reality is that fiber is not a economically viable option for many areas due to geography and demographics, and yet quite viable for others. For the average consumer, <100ms and >50Mbit is more than enough for non-competitive gaming, streaming, working from home, etc. For those whose options are limited, or simply not as good (whether performance, price, or both), Starlink is a great option. For those that it can't compete with the established ISPs - that's fine too. Starlink doesn't need to, and realistically technically can't, attempt major market saturation in denser areas where such competitive ISPs would be.

 

[banned-66611]

Most fiber networks are not following the shortest path, but weird paths along various rights of way (such as old rail lines, or other utilities). So other than last mile you can pretty much guarantee it'll never be shortest path.

Sure, but Starlink also isn't the shorted path as the satellites are in orbit, meaning the signal has to go up 340km and then down 340km, or further if the satellite isn't directly over both the user and the base station. 680km combined before you add any lateral distance.