3 ms·

I mean I think imagining it as 100GB of high priority traffic per day is the wrong way to think about it. I know that in some places in the US for some reason y

I mean I think imagining it as 100GB of high priority traffic per day is the wrong way to think about it. I know that in some places in the US for some reason you pay for the bandwidth you use, instead of for the capacity of your line. Personally I think that's silly. With a 100Mbits/s connection you'd have 10Mbits/s of high priority traffic, per second. I feel like an ISP should be able to not completely oversubscribe to the point where you can't even use 10% of your connection at low-latency.

dtaht 5y ago
by multiplexing better, you generally need to not differentiate between high and low priority. Being that I've been trying to convince people of this for years - all I can do is try to point at resources that demonstrate how we think about how packets work is frequently wrong, and/or try to provide a demonstrable example. Let's take voip vs netflix over 60ms - netflix, like most DASH video traffic from most video services, is very bursty. It tries to grab a chunk of video, and over the course of 1-2 seconds, fills the pipe, and voip sends one tiny packet every 20ms, which kind of looks like this: What you want: NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN -> you V V V What you get: VVVNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN -> you What AQM does is keep that induced queue short (roughly 16ms with DOCSIS-PIE). The netflix "pulse" goes away and you end up with NNNNNNNNNNNVNNNNNNNNNNNNNN -> you Netflix loads the next segment a tiny fraction slower, but your voip call doesn't get the jitter and latency side-effect anymore.
dtaht 5y ago
I've given more than a few lectures on how packets and queues and aqm are supposed to work, with very physical demos, using "packets as people", such as this one: https://blog.apnic.net/2020/01/22/bufferbloat-may-be-solved-but-its-not-over-yet/ https://blog.apnic.net/2020/01/22/bufferbloat-may-be-solved-... I'm cited in the article, participated in the study, and have been trying to counter this mis-impression about "priority" vs filling the pipes, not the queues, as the answer to better network latency for many, many years now. It's still a tough slog, it seems, and I fear this will be a long day on reddit for me! But with comcast's now enormous and deployed existence proof, perhaps more will deeply grok it, and we'll see movement by other isps to apply the same technologies so all their users would benefit.

dtaht 5y ago

by multiplexing better, you generally need to not differentiate between high and low priority. Being that I've been trying to convince people of this for years - all I can do is try to point at resources that demonstrate how we think about how packets work is frequently wrong, and/or try to provide a demonstrable example. Let's take voip vs netflix over 60ms - netflix, like most DASH video traffic from most video services, is very bursty. It tries to grab a chunk of video, and over the course of 1-2 seconds, fills the pipe, and voip sends one tiny packet every 20ms, which kind of looks like this: What you want: NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN -> you V V V What you get: VVVNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN -> you What AQM does is keep that induced queue short (roughly 16ms with DOCSIS-PIE). The netflix "pulse" goes away and you end up with NNNNNNNNNNNVNNNNNNNNNNNNNN -> you Netflix loads the next segment a tiny fraction slower, but your voip call doesn't get the jitter and latency side-effect anymore.

dtaht 5y ago

I've given more than a few lectures on how packets and queues and aqm are supposed to work, with very physical demos, using "packets as people", such as this one: https://blog.apnic.net/2020/01/22/bufferbloat-may-be-solved-but-its-not-over-yet/ https://blog.apnic.net/2020/01/22/bufferbloat-may-be-solved-... I'm cited in the article, participated in the study, and have been trying to counter this mis-impression about "priority" vs filling the pipes, not the queues, as the answer to better network latency for many, many years now. It's still a tough slog, it seems, and I fear this will be a long day on reddit for me! But with comcast's now enormous and deployed existence proof, perhaps more will deeply grok it, and we'll see movement by other isps to apply the same technologies so all their users would benefit.