Simple, it makes no sense to define a latency for non-interactive stuff.
No it doesn't. My input affects my car. If the input has to be transmitted to a server, that'll add 10ms latency.
The physics modelling is all done server side, tyres collision, the works. The result is transmitted back to my console, another 10-15 ms of latency.
Add this together and this is the roundtrip latency. What are you trying to discuss?
What does 10ms means? Assuming you drive a sports car with decent tyres, you can brake and accelerate around 1G (10m/s/s) that'll give you change in velocity of .36km/h or .23 mph, ie very little.
You have seen that I wrote that such a delay is usually quite tolerable (I was even mentioning some higher numbers)?
No you don't. You only need to predict the one-way latency. The console is a one-way transmission delay behind the server.
And the server is a one-way transmission delay behind the console.
If we talk latency, it means the (additional) delay between some event (user input, some simulation result/collision detection on the server) and when the possible reactions to this returns from the other side (some information has made the roundtrip). It simply doesn't work any other way, you know, causality and stuff.
Simple example: Some alien race on a planet 5 lightyears away has the highest power supercomputer in the universe which can handle all conceivable problems in basically no time. We send them some problems to runs and they act according to our instructions. What is the perceived latency for all situations involving human feedback to something running there is needed? Answer: 10 years.
If they ask us a question, our answer arrives there 10 years later. If we ask them a question, their answer arrives here 10 years later.
To sum it up: Every communication involving feedback (i.e. interactive) has always at least the roundtrip latency. How much problems arrise from that fact depends on the scenario of course. As I said, for racing games a latency of several 10s of ms should be tolerable.
