Earlier quoted context omitted.
> In other words, network topology adapts to traffic demand (albeit more slowly than traffic shaping). A fair traffic management policy would allow for YouTube users to consume bandwidth in exact proportion to their share of overall demand for bandwidth. Traffic shaping on top of that could only have the effect of meaning that a YouTube user will get more bandwidth than their Netflix-using neighbor even if they're bo…
> A fair traffic management policy would allow for YouTube users to consume bandwidth in exact proportion to their share of overall demand for bandwidth. I think this statement is in effect a straw man, since the relevant consideration for QoS is the timing of consuming that bandwidth. If you mean instantaneous bandwidth, then you are saying that both Youtube customers and Netflix customers should be able to use a ci…
> If you argue that a VOIP customer should be able to place a call with the absolute minimal latency the network can provide, even if it means that video content will (on average) buffer for an extra second, that is not an example of a fair system, just one that is extremely friendly to VOIP."
That's the real straw man. The current state of the art for QoS is active queue management that both increases effective link utilization and reduces latency to a point that VoIP is usable, as compared with naive FIFO queuing with tail drops when the queues are full. That's why I said QoS isn't a zero-sum game. In the real world, you don't actually have to choose between throughput and latency; both can be good enough for what people normally want to do. VoIP traffic doesn't cost enough bandwidth to have a significant impact on Netflix video quality, and firing up a Netflix stream doesn't have to produce queuing latencies that make VoIP unusable. And none of this requires any traffic shaping rules that deliberately identify what's video or VoIP—that's an objective criteria for "neutral" that doesn't force ISPs to use naive sub-optimal traffic management.