> Latency is the issue in routers when they categorize traffic, that was the idea.
That's what I'm contesting.
Suppose you have a router which is receiving data to be sent over a 1Gbps link. Over a one hour period there is an average of 600Mbps of streaming video, 400Mbps HTTP traffic and 300Mbps of FTP/BitTorrent/etc. traffic. There is more data than there is link to put it in; you are screwed. The streaming video is going to stutter or degrade to lower quality, web pages will be slow to load, people waiting for downloads to finish will have to wait longer, because there is not enough bandwidth. The latency will also be poor, if router buffers are set too large and create a large queue in front of new packets (bufferbloat), but that's the least of your worries in that situation. Fixing the latency wouldn't clear things up because you still have users trying to play video with a bitrate of 6Mbps through a connection with a 4Mbps throughput. And categorizing the traffic only changes who gets screwed -- if you put downloads at the bottom of the heap then you might make the streaming video and HTTP customers happier but the customer who paid for a 6Mbps connection and would have at least been downloading at 4Mbps is now waiting for a download which getting only 0.2Mbps. That's not a solution, it's triage.
Now suppose you have the same amount of traffic on average but the link is upgraded to 2Gbps. Now there is no packet loss. You might have short bursts where the traffic level exceeds the capacity of the link, but they get smoothed out by the router's buffers, which never get full so the queue length is never more than a few milliseconds. Solving the bandwidth problem solves the latency problem.
The point is, if you're trying to fix the latency that occurs as a result of router buffers getting full or dropping packets in the core of the internet, you're rearranging the deck chairs on the Titanic. The underlying problem is that router buffers are getting full or dropping packets in the core of the internet.
That's what I'm contesting.
Suppose you have a router which is receiving data to be sent over a 1Gbps link. Over a one hour period there is an average of 600Mbps of streaming video, 400Mbps HTTP traffic and 300Mbps of FTP/BitTorrent/etc. traffic. There is more data than there is link to put it in; you are screwed. The streaming video is going to stutter or degrade to lower quality, web pages will be slow to load, people waiting for downloads to finish will have to wait longer, because there is not enough bandwidth. The latency will also be poor, if router buffers are set too large and create a large queue in front of new packets (bufferbloat), but that's the least of your worries in that situation. Fixing the latency wouldn't clear things up because you still have users trying to play video with a bitrate of 6Mbps through a connection with a 4Mbps throughput. And categorizing the traffic only changes who gets screwed -- if you put downloads at the bottom of the heap then you might make the streaming video and HTTP customers happier but the customer who paid for a 6Mbps connection and would have at least been downloading at 4Mbps is now waiting for a download which getting only 0.2Mbps. That's not a solution, it's triage.
Now suppose you have the same amount of traffic on average but the link is upgraded to 2Gbps. Now there is no packet loss. You might have short bursts where the traffic level exceeds the capacity of the link, but they get smoothed out by the router's buffers, which never get full so the queue length is never more than a few milliseconds. Solving the bandwidth problem solves the latency problem.
The point is, if you're trying to fix the latency that occurs as a result of router buffers getting full or dropping packets in the core of the internet, you're rearranging the deck chairs on the Titanic. The underlying problem is that router buffers are getting full or dropping packets in the core of the internet.