Closed cpaasch closed 2 years ago
Groovy.
There are other ways to calculate the saved value of reducing latencies on the internet that someone could explore that result in much bigger and ego-swelling value, that I try not to think about much. I'm content with 1sec a day!!, but...
Take drops/day for example. On the network that I used to maintain at the campground, those were about 3000-9000/day (about 5% marks also) for 3-30 active users, and I knew each of those was preventing a 600+ms (or 240+ms) latency excursion that might have lasted for seconds or minutes for all the other users, and a network that pretty consistently carried voip and videoconferencing well (except on the farthest reaches of the wifi).
A number I keep in my head is loading slashdot took 4 minutes with 1sec latency, 14s with none.
Reschedules were much, much higher, (and not a particularly accurate measure) so every time fq is invoked you save the time of the length of the rest of the queue.
Then of course you have the enormous apple deployment of fq_"codel", interleaving away... the linux cloud deployment, the chromebook deployment....
It would be kind of cool to somehow total up an estimate of how valuable the whole bufferbloat.net effort has been to the internet as a whole. What if jim had NOT looked up from his scp to MIT that day and started asking questions?
anyway, it's somehow expressing that the routers are the only thing major left to upgrade, rather than me ranting about millions and billions of fixes elsewhere, that was more on my mind.
Feedback from Dave Taht: