dtaht

@NetdumaFraser I note I know absolutely nothing about netduma's anti-bufferbloat solution, although I am under the impression it uses fq_codel, at least in part. (?) Is how it works documented? Is it derived from the sqm-scripts? I did want to call your community's attention to the latest effort of the bufferbloat project, "sch_cake", which solves a few problems, like per-host fq, ack-filtering, docsis framing and so on. A paper describing some of its features is here: https://arxiv.org/pdf/1804.07617.pdf We describe in more detail what it solves here: https://www.bufferbloat.net/projects/codel/wiki/CakeTechnical/ It is available as a backport to many linux versions and was just accepted into linux net-next which means it will be part of linux 4.19. On the way to getting it mainlined (and underdocumented at present) are means to keep you dpi working even more sanely than it presently does. I doubt your offloaded nat system can be made to work for per-host fair queuing implementation (which we think is the bomb), but what do I know? Happy debloating! -- Dave Taht co-founder bufferbloat project