Channel congestion in dense urban areas approaches the same limit, heck Ars was just talking about a technique today[1] that because it doesn't respect channel slots drops throughput of any router within 60m by 20%.
100-200ms is an upper bound, however they were reliably doing client-side latency prediction back in '96[2] over 300-400ms dial up connections with little fidelity loss including servers supporting 200+ concurrent players.
I'm sure there's a niche of people for which streaming is usable, however we would design games with latency in mind(and how we'd deal with it) as a part of fundamental mechanics of the game. Here they're just slapping them on a video stream and hoping it works.
[1] https://arstechnica.com/gadgets/2019/11/where-the-wi-fi-suck...
Your point still stands though, on a wireless network, it only takes some interference from other devices to cause latency spikes that make latency sensitive games frustrating to play, even if the actual upstream connection is of high quality. I've had quite a bit of issues with playing games on wireless networks in urban areas that go away magically by using an ethernet cable; the stuttering that happens when ping spikes from 30ms to 400ms is very jarring.
If Stadia is used for games that are not latency sensitive, such as deckbuilding games or turn based strategy games, it's fine. Using it for competitive first person shooters is likely to be very frustrating on anything but a wired network with good upstream connectivity or a wireless network with no major noise sources.
At the end of the day the internet is a packet switched and not circuit switched topology despite telecom pushing hard for circuit switched in the early days. Preference is given for throughput and not latency(although network-next is working to address some of that).
In my mind things like Stadia will increase the amount of data traffic exponentially, even at a low market penetration rate. I think I saw one estimate that it's going to be 120gb on average per hour. Even at 100gb, the extra traffic at scale is going to be insane.
Regardless of if current infrastructure can handle x% market penetration of that kind of usage in a, what, 1 year time? Should the infrastructure even be required for that kind of insanity? We're not talking about life-saving data transportation. We're talking about god damn video games. Or am I way off my whiskey fueled rocker?
This estimate is waaaaay off. Nobody has been shipping video encoders or decoders setup to pull 1000fps nor tuned for latency.
Realistically it's 8-14ms to decode if you're lucky (low latency hardware+firmware), and comparable to encode. And these streams are made entirely of key-frames, which is a hyper unusual scenario for video systems.
That all adds up to maybe something like +30ms over baseline. Which is a lot. If you have a console try playing it on a TV with and without game mode enabled. That's about a 20-30ms latency difference in general.
Well, I'm in one of the 10 largest cities in the United States. Pinging 8.8.8.8 shows an average of 20 ms RTT, which is just over one frame. So is the total a 2 frame delay? No. Think about how Youtube works: they choose to introduce more latency than exists in the raw connection for reliability. My ping time shows a std deviation of over 10ms to 8.8.8.8. So if Youtube ran at a 20 ms latency (the true network latency), just about every other frame would have to briefly pause to wait for the next frame to arrive over the network. So you introduce latency. Youtube introduces about 3 seconds of latency on my network (a very rough estimate).
With a video game, the situation can't be fixed like that. You can't just run your video game behind your inputs by 3 seconds to iron out network latency spikes. So you've got to pick a percentage of packets that are going to arrive late (briefly pausing the stream), and just hope that you never get huge latency spikes of a second or so. (This actually happens all the time on busy networks due to buffer bloat.)
If you pick 95%, 1 out of every 20 video segments are going to have to pause for the network to catch up. Given we're expecting 60fps video, I'm assuming that's considered unplayable. So lets make it 99%. Assuming latency follows a bell curve, that means you have to introduce 45 ms of latency to allow smooth playback. That rounds to 3 frames of latency.
So all together, the "ideal" experience in my major city is to play a video game with 4-5 frames of latency, and realistically my developers are probably asking for a 100ms input-to-paint window. That's TERRIBLE to the point that it's probably only really usable for "clicker" time-waste games and turn based stuff. And the thing is, latency doesn't follow a bell curve. In my test lasting less than 30 seconds, I saw ping times well over 60 ms, and hardware associated with gaming will have its own latency issues too. So I expect even a stream running at 100 ms latency will still have brief pauses and jumps every few seconds, and probably noticeable ones every few minutes.