So let's think about how traffic gets onto the network and what steps might make sense to limit that. I have some "crazy" ideas about this including per device reputation enforced as close to the device as possible. Yes, if we say that anyone with any sort of device can send data to anyone then this will be a problem. There are other options including different sorts of "darknet" type things. Are there no "outside the box" type solutions that you can think through the tradeoffs for? I think the underlying assumption you're working with, that anyone anywhere on the network should be able to drop an unlimited amount of data onto the link headed to me as a rule of how things must forever work needs to be justified.
Yes, that would be a valid solution - authenticate every device, or provide a per-device reputation. But this has a couple of problems that I can think of:
1. Per-device reputation removes the concept of anonymity. If I can look up the "reputation" of the device that sent me a packet, I can track it perfectly too.
2. Authenticating every device (beside the practical challenges) is very inconvenient. What happens if I move countries? Buy a new phone? Or a new network card?
And there's more issues that I won't list :)
Problems aside, I agree with the statement: "the underlying assumption [...] that anyone anywhere on the network should be able to drop an unlimited amount of data onto the link headed to me [...] needs to be justified".
I think the most practical solution to this would simply be forcing ISPs (through legislation would be best) to look a little closer at their traffic. If I'm running an ISP, and I see a computer making 100 requests/second to a single website for more than a minute, I'm immediately thinking "DDOS". Yes, there's privacy issues, but most ISPs already do some sort of traffic shaping (see: Sandvine), so it shouldn't be that much of a stretch.
It'd need the help of browsers or OS's (depending on where in the stack you put the logic), but one idea might be to require requests/packets to be signed by something that proves a sufficient amount of CPU work has been done (ala bitcoin). If the site comes under attack, they could turn this on (presumably with a middle-man service that can take high bandwidth) and up the amount of work required to reach the destination. This would no doubt slow things for the legitimate users, but it could make things much more difficult for the attackers.
This really doesn't solve the DDOS problem though. It's throwing more CPU time and bandwidth at a scenario that already requires both of those. It can slow a small group of script kiddies making a thousand requests to your server per second, but it doesn't stop an actual distributed attack using a botnet or large numbers of machines.
If you're adding the signatures, you presumably need to spend CPU time to authenticate it, and bandwidth to send the data, plus the actual content. Why not just have the middle-man soak up the extra requests, cache the data, and fan it out that way?
