But for something more realistic for gaming, a player in Boston hitting a server in NY. The fiber message travels something like 200 miles, while the satellite message travels 2,200 miles.

Radio waves are faster than light going through a fiber optic cable, and the satellite mesh network will have fewer hops (switching packets on fiber is expensive time-wise).

The difference in latency won't be nearly as bad as a naive comparison of these two numbers would suggest. Depending on what hops your terrestrial traffic is going through, the satellite transmission could well be faster. To give you a trivial example, I'm seeing 11 ms ping right now to a server within my own city, that's only a couple miles away. This is ~0% signal transmission delay and ~100% switching delay.

It's still transmission delay, you are underestimating how fast routers forward packets today. Likely the server has a different provider and them and your ISP does not peer in your city, so you have to go somewhere else where they peer (or where their transit peers).

I think no ISP would connect to others with a huge coil of fiber to add that amount of latency.

No? But they do not connect to each other in every city either, which is why you can get this delay even if both parties are in the same city.

All the uplinks of my ISP with one exception (carrying traffic through various lines up to european IXs) are connected to the Internet through the same MSK-IX which is 900 miles away from my city and adds 4ms latency. I don't think GGP's traffic route is that long though.

Not sure what you are talking about. 900 miles is ~1500km, 1500km at the speed of light in fiber (2/3 of SOL) takes 7,5 milliseconds, so 15ms roundtrip.

But I don't see why it's relevant. Forwarding in modern ISP routers is on the tens to hundreds of microseconds level, even if there are many hops it does not add up to that much. So what he has is likely not "switching delay", and even saying that 11ms is 100% switching delay when it's the same city (low amount of hops even if multiple ISP's are involved) is just ridiculous.

> takes 7,5 milliseconds, so 15ms roundtrip.

Took that number (it was for one way trip though) from memory, might be scrambled. Just checked, RTT 17ms with 3ms jitter for me.

22ms RTT is still too much though. Even with centralized network for cross-border traffic like in my country, there are IXs way closer than those 1500km to any town on european part and obviously to the East of Urals (the longest distance between IXPs is ~3000km for Krasnoyarsk-Khabarovsk, the longest distance to the closest IXP of 1500km have Norilsk - the most northern city on the globe and it's satellite Dudinka).

edit: fixed some english.

So maybe satellites simply aren't the best-use case for gaming?

I don't think Elon is counting on gaming to be the backbone of this service.

Boston to New York is 215mi or a 430 mile round trip. At 200 miles up where they intend to put the satellites orbiting over major cities, it's going to be about 1000 mile round trip.

I thought that the US servers for most games are in the middle of the country to get the best average ping for a cross-country audience. Blizzard is like Chicago and Texas.

> Boston to New York is 215mi or a 430 mile round trip.

Assuming terrestrial lines travel as the crow flies, which is not all that likely (but probably not too far off).

> At 200 miles up where they intend to put the satellites orbiting over major cities, it's going to be about 1000 mile round trip.

Other people here are noting that fiber signals travel at ~70% the speed of light, while radio waves travel at very close to the speed of light. If that's true, it puts the optimal time for land based optical communication at ~3.3ms[1] ms, while the optimal time for a low orbit satellite at ~5.4ms[2] (if I didn't screw up the math). That's slower, but not by a whole lot, so it might even be better than you're predicting.

Also, are people actually getting sub 4ms round trip times between Boston and New York? If not, then the difference here might be entirely subsumed by other factors.

1: 430 Miles / 186,000 M/s / 0.70 = 0.0033 s

2: 1000 Miles / 186,000 Miles/s = 0.0054 s

Nope, the biggest AWS region is in Virginia (as well as a bunch of others, energy is cheap there). There are also lots of datacenters in Dallas, TX and San Jose, CA. NYC has a lot of connectivity (throughout the northeast, and transatlantic) but the cost of real estate is so high that it is uneconomical to operate lots of large datacenters there.

The majority of the population in the US is in the eastern half (1/3 in states on the East coast), so you don't actually decrease median latency by putting datacenters in the geographic center: https://commons.wikimedia.org/wiki/File:United_States_popula...

Nope, the mean population center of the contiguous United States is somewhere west of St. Louis.

https://en.wikipedia.org/wiki/Mean_center_of_the_United_Stat...

Actually not too far away from the geographic center which is somewhere in Kansas.

That doesn't disagree with what I said. The majority is still in the East (the mean is East of the geographic center), and you don't want to optimize for average latency anyway. You'll get the best median latency with two datacenters by putting one on the mid-Atlantic coast (~180M people East of the Mississippi) and one on the mid-Pacific coast (another 50M in Pacific Coast states). That means the latency for ~2/3 of the population is way lower, and the max latency is still about the same as putting them in the center of the country.

Not usually. In my experience, they usually follow the typical popular datacenter areas e.g. East Coast/West Coast. I've mostly only seen really popular games offer Central locations like Dallas/Chicago. But it's getting a lot better now that vultr (COD uses them) and other VPS providers get more and more locations.