"All incoming traffic is going to end up with the QoS flag set, and thus the internet will basically behave as if it didn't exist."
Except people will be expected to pay a premium for it. Ie, it's a glorified rate hike masked as a technical non-sequitur.
TL;DR: QoS may be useful for networking, but it is not useful for inter-networking.
The only use for QoS IMHO is an "emergency lane" for health-and-safety traffic, assuming it carried the same disincentives to abuse (jail) as does 911 and other emergency services. (Not even VoIP is "high-priority" enough to warrant it.)
The other problem with the "sender pays" model is that there's no basis for it. You're not "tying up the line" by sending any more than you are by receiving. You are either online or you aren't. Anything else is rolling backwards downhill, away from the Internet.
I live in Canada, and all of the ISPs wires were subsidized with taxes then handed over to the incumbent carriers to use and sub-let.
A network can saturate, that's fine. I have a link that's either on or off, that's fine too. If the overall network speed is disatisfactory, that means it's time to upgrade the switches -- possibly at my expense (I presume that's what I've been paying for) -- but the Internet has always been best-effort and that's also the point. It automatically load balances. QoS is a solution in search of a problem. The solution has always been at the switch when it's saturated. A CPU process scheduler may boost a process temporarily to increase responsiveness, but only for an instant. The long-term, steady state, is still a free-for all. Nothing stops you from having QoS on your network or PC, but it is not a core service.
ISPs simply need to stop over-selling rates and attempting to "manage" link speed to compensate. Stop saying there's a difference between plan A and plan B if they come over the same wire. There's no basis for telling subscriber A that they get 512Kbps when subscriber B get 3Mbps over the same wire and there's no technical difference... They both have that wire and that wire is paid for, or being paid for, by their subscription. Both subscriber A and subscriber B pay the same rate and compete for the same throughput. The switch balances throughput fairly and it costs the ISP no more than ever. I'm not against creative pricing and free markets, just against single-use networks being called "the Internet." QoS may be a networking technology but it is not an Internet technology.
The whole ISP pricing model is artificial, and moving towards the grossly-overpriced text message pricing. The only reason small ISPs can't offer anything like I have described (anymore) is because they have no choice but to buy service through incumbents who are inflicting these QoS and bandwidth management fictions.
I have run a network for thousands of users (academic; students with torrents competing with instructors with videoconferencing, and administrators with VoIP), and while it may be appropriate to classify traffic (ex, emergency, or non-emergency), when the customer is "the public" and the context is "a democracy" all traffic must be treated equally in the extreme, otherwise it is simply discrimination. Even if you classify traffic as "government" and "non-government", or "porn" and "non-porn" you have big social fall-out.
"The Internet" is best-effort by design and that is more than enough. In my experience well-intentioned administrators without real-world experience come in and try to "solve" the latency problem creating more problems. Latency needs to be fixed at the switch in the same way CPU scheduling is: if a link hasn't spoken recently, it gets a little short-lived priority boost. The user has no say.
These management decisions are coming from the same people who buy "smart" switches that cost 10 times as much so they can use VLANs as a security technique, when it isn't, and when PPPoE would have worked just as well and provided the desired encryption.
We consistently invent little nested versions of our universe (even going so far as to tunnel it all over HTTP, because that's the only port that anyone leaves open now), and we wonder why it's slow and expensive and flaky, and we're no further ahead than we were when we started.
Except people will be expected to pay a premium for it. Ie, it's a glorified rate hike masked as a technical non-sequitur.
TL;DR: QoS may be useful for networking, but it is not useful for inter-networking.
The only use for QoS IMHO is an "emergency lane" for health-and-safety traffic, assuming it carried the same disincentives to abuse (jail) as does 911 and other emergency services. (Not even VoIP is "high-priority" enough to warrant it.)
The other problem with the "sender pays" model is that there's no basis for it. You're not "tying up the line" by sending any more than you are by receiving. You are either online or you aren't. Anything else is rolling backwards downhill, away from the Internet.
I live in Canada, and all of the ISPs wires were subsidized with taxes then handed over to the incumbent carriers to use and sub-let.
A network can saturate, that's fine. I have a link that's either on or off, that's fine too. If the overall network speed is disatisfactory, that means it's time to upgrade the switches -- possibly at my expense (I presume that's what I've been paying for) -- but the Internet has always been best-effort and that's also the point. It automatically load balances. QoS is a solution in search of a problem. The solution has always been at the switch when it's saturated. A CPU process scheduler may boost a process temporarily to increase responsiveness, but only for an instant. The long-term, steady state, is still a free-for all. Nothing stops you from having QoS on your network or PC, but it is not a core service.
ISPs simply need to stop over-selling rates and attempting to "manage" link speed to compensate. Stop saying there's a difference between plan A and plan B if they come over the same wire. There's no basis for telling subscriber A that they get 512Kbps when subscriber B get 3Mbps over the same wire and there's no technical difference... They both have that wire and that wire is paid for, or being paid for, by their subscription. Both subscriber A and subscriber B pay the same rate and compete for the same throughput. The switch balances throughput fairly and it costs the ISP no more than ever. I'm not against creative pricing and free markets, just against single-use networks being called "the Internet." QoS may be a networking technology but it is not an Internet technology.
The whole ISP pricing model is artificial, and moving towards the grossly-overpriced text message pricing. The only reason small ISPs can't offer anything like I have described (anymore) is because they have no choice but to buy service through incumbents who are inflicting these QoS and bandwidth management fictions.
I have run a network for thousands of users (academic; students with torrents competing with instructors with videoconferencing, and administrators with VoIP), and while it may be appropriate to classify traffic (ex, emergency, or non-emergency), when the customer is "the public" and the context is "a democracy" all traffic must be treated equally in the extreme, otherwise it is simply discrimination. Even if you classify traffic as "government" and "non-government", or "porn" and "non-porn" you have big social fall-out.
"The Internet" is best-effort by design and that is more than enough. In my experience well-intentioned administrators without real-world experience come in and try to "solve" the latency problem creating more problems. Latency needs to be fixed at the switch in the same way CPU scheduling is: if a link hasn't spoken recently, it gets a little short-lived priority boost. The user has no say.
These management decisions are coming from the same people who buy "smart" switches that cost 10 times as much so they can use VLANs as a security technique, when it isn't, and when PPPoE would have worked just as well and provided the desired encryption.
We consistently invent little nested versions of our universe (even going so far as to tunnel it all over HTTP, because that's the only port that anyone leaves open now), and we wonder why it's slow and expensive and flaky, and we're no further ahead than we were when we started.
/rant