zlacker

So which points are getting “faster than fibre” latency because of this? Extra distance up and down, but make up for it on the long-haul.

Won’t beat HF radio though.

>>Scound+(OP)
Why would we expect faster than fibre?

>>edgyqu+j3
The speed of light in a vacuum is roughly 50% higher than the speed of light in fiber.

>>Scound+(OP)
That could already be the case. Round trip time to the ~500km orbit is about 4 milliseconds (+ all other network elements before, after and in between). They claim to have a >5000km link running for significant time. Now think of a fibre link of that length and how many repeaters / routers will be needed due to attenuation and physical constraints. I can clearly see a path where Starlink laser links could be a viable option to subsea cables - at least for some priority traffic...

>>Cu3PO4+k4
Is attenuation in a vacuum also better?

>>Scound+(OP)
Optical fiber has an index of refraction of around 1.6, so signals travel at around 0.6c. For a perfectly straight cross-continental link (5,000km) with no delays from amplification/retransmission, that's about 26 milliseconds. Assuming the satellites are directly overhead, Starlink adds another 500km up and down, making the minimum possible latency around 20 milliseconds. The real number might be slightly higher or lower depending on the location of the satellites.

My guess is the real latency depends mostly on the latency of relay nodes (either satellites or routers on earth), not the medium through which signals travel.

>>oger+B4
a few random outages which happened near places some oligarch's yacht has visited recently and it'll become the priority backhaul.

I see folks in the Pentagon doing a collective /phew that this project is online in the next decade, multiple times.

>>ggreer+f7
Number of hops definitely matters more usually. For example I'm about 150 miles from Azure East US 2 (richmond, va), and at the speed of light that should be sub 2ms round trip, but actual latency to it is ~30ms. But I'm sure I'm going through dozens of switches/routers to get there. What Starlink buys you is that you get to go straight to a satellite, then a laser in a vacuum to other satellite(s) and then a ground station that's likely already at an IXP or very close to one.

>>edgyqu+j3
Also, as pointed out elsewhere, the number of hops is the biggest contributor to latency.

>>andrew+I6
yes, glass has a non zero amount of absorption which is why Erbium amplification is required.

>>Cu3PO4+k4
The area between a starlink receiver on the ground and a satellite isn’t a vacuum

>>edgyqu+kl
That's a great point, I was curious so I looked it up. Google offered the following:

"The speed of light in air is about 299,705 kilometers per second, or 2.99705 × 10^8 meters per second. This is almost as fast as light travels in a vacuum, slowing down by only three ten-thousandths of the speed of light."

So seems like the speed of light in atmosphere is still a lot faster than fiber.

>>edgyqu+kl
The speed of light in air is 99.97% of that in a vacuum, vs about 2/3 of c through fiber.

>>Scound+(OP)
It's possible for starlink to beat radio, because radio can't always go straight to the target. If I wanted a radio link from NY->Tokyo, what would that path look like?

>>oh_sig+io
True, it would be bouncing around between ground and the atmosphere (when it works at all).

>>Scound+(OP)
Starlink adds a latency penalty of tens of milliseconds going through the atmosphere. Each round trip is four hops through the clouds. I expect most of this delay is forward error correction, combined with lower bandwidth of the radios.

On top of that, you may have queuing in each satellite.

Finally, the satellite laser links aren’t pointing exactly in the direction you want to your packets to travel. They’re at some diagonal, and the packets need to tack back and forth, which wastes distance. Think the streets of Manhattan.

>>Cu3PO4+k4
I will note that this is the case for conventional fiber-optic cable. The newer hollow-core fiber cables transmit light at nearly c. As far as I know hollow-core has not seen wide-spread use, but it will be interesting if trans-continental connections switch over.

>>oh_sig+io
It would look like the HF radio bouncing off the ionosphere. I have contacted someone in Japan from Oregon. The downside of HF is that the bandwidth is low with 30MHz across the entire band.

There was company recently wanting to do high-frequency trading on HF because of the quickest path.

>>minhaz+Xc
Could also be shit routing.

Some big ISPs here refused to locally peer with some cheaper providers, so some packets to a local data centre (5 miles away) in Toronto would round trip through Chicago and back.

If they wanted a direct connection; they wanted them to pay for transit.

>>ortusd+9B
Will adopting the new cables everywhere mean nearly halving latency all across the board? Sans routing latencies.

This would be huge for realtime gaming across continents.

>>ianbur+4I
I'm curious how much the curved route affects the effective speed of communication with skywaves. You also have many situations where communication becomes impossible due to space weather or other atmospheric phenomena.

>>ByThyG+5N
Any stretch of fibre you replace with hollow core fibre will see latency reduced to two thirds of what it was before. (It would be half if the speed of light in it were double what it is in normal fibre, but it's only 50% faster).

You say sans routing latencies, but these are very much significant for intercontinental communication:

I get 6ms ping to AWS eu-central, which is less than 100km by air from me. I get 114ms to AWS us-east-1, which is roughly 6500km. Now 6500km / (2/3 * c) = ~32ms. So if there were a fibre running in a straight line, time in the fibre would be 32ms. Of course it isn't running in a straight line, so let's say 50ms are pure "light traveling through fibre". Switching all of that to hollow-core would cut that to 33ms, so that's a savings of 17ms or roughly 15% of my total latency.

This is still a very nice savings, but very far off from cutting latency in half.

(Also, it's a single hop from my company network to DE-CIX, one of the largest internet exchanges in the world, so I feel confident saying my results aren't skewed by a bad uplink.)

>>7e+zr
This is just incorrect. The speed of light through atmosphere is almost identical to speed of light in a vacuum. There's no latency penalty for traveling through the atmosphere. The one-way time delay to a Starlink satellite is about 2 milliseconds.