On Tue, Jun 7, 2022 at 8:24 AM Livingood, Jason via NANOG <nanog@nanog.org> wrote:
A related observation – years ago we gave cable modem bootfiles to a group of customers that had no rate shaping according to their subscription and compared that to existing customers (with an academic researcher). The experiment group did not know of the change, so it could not influence their behavior. We observed that peak demand generally hit a plateau that was well below available capacity & this was driven by existing applications & associated user behavior. There’s obviously a chicken-or-egg problem with capacity & apps to use that capacity, but most ISPs raise end user speeds at least annually and try to stay ahead of increases in peak demand.
I think peak demand should be flattening in the past year? There's only so much 4k video to consume, so many big games to download? My curve seems closer to a doubling of the average usage over 10 years. It would be really radical of me to start yelling "peak bandwidth" a la peak oil without more study... A very informal survey of those that had deployed higher rates on mikrotik stuff at WISPAMERICA had all 5 of the people rolling their eyes and saying avg downloads had gone from 2 to 3Mbit upon doubling or more their allocated bandwidth, and they had no congestive issues on their network peering. There was also a technical limitation in the mikrotik deployment in that they use very short queues by default (50 packets) for either the fifo or (the common) SFQ deployments. Shapers were universally used by this small group, and they were unaware of the sideffects of such short queues. I also took apart a recent ubnt 60Ghz radio's behaviors, and that was FQ'd and also with very short queues... and what looked like ack synthesis... with no options to change the configuration. I am thinking in part the lack of measured WISP "demand" for more bandwidth is in part due to overly short (as opposed to bufferbloated) queues! There's a really long thread over here with the mikrotik userbase going to town on fq_codel and cake: https://forum.mikrotik.com/viewtopic.php?p=937633#p925485
JL
From: NANOG <nanog-bounces+jason_livingood=cable.comcast.com@nanog.org> on behalf of Jim Troutman <jamesltroutman@gmail.com> Date: Monday, June 6, 2022 at 19:29 To: Tony Wicks <tony@wicks.co.nz> Cc: "nanog@nanog.org" <nanog@nanog.org> Subject: Re: FCC proposes higher speed goals (100/20 Mbps) for USF providers
Some usage data:
On a rural FTTX XGS-PON network with primarily 1Gig symmetric customers, I see about 1.5mbit/customer average inbound across 7 days, peaks at about 10mbit/customer, with 1 minute polling. Zero congestion in middle mile, transit or peering.
-- FQ World Domination pending: https://blog.cerowrt.org/post/state_of_fq_codel/ Dave Täht CEO, TekLibre, LLC