I have couple of machines where I've setup PF in following way: pce-0041# grep -w queue /etc/pf.conf queue hfsc_fq on em0 flows 1024 qlimit 200 quantum 300 bandwidth 85M max 85M default queue wifi_fq on athn0 flows 1024 qlimit 100 quantum 300 default
pce-0035# grep -w queue /etc/pf.conf queue q_umb0 on umb0 flows 1024 qlimit 50 quantum 300 default queue q_athn0 on athn0 flows 1024 qlimit 100 quantum 300 default pce-0067# grep -w queue /etc/pf.conf queue q_athn0 on athn0 flows 1024 qlimit 100 quantum 300 default I'm wondering does it make any sense to have fq_codel without hfsc, in general (no matter is it wireless or wired interface) and especially in case of wireless cards (LTE or Wi-Fi)? Above machine pce-0035 which has umb(4) during the day has terrible network performance to the internet. I think it is just the location, because during nights (after midnight and before 7am local time, approx) speedtest-cli shows order of magnitude beter results (~12Mbit/s at night versus ~0.4Mbit/s during the day). Users of LTE network in that location complain about the internet all the time, no matter is it their Android phones and their data plan or via that OpenBSD access point via athn(4) via umb(4). I'm just wondering am I making things worse by introducing queuing on that type of interfaces? Is this very empirical problem and I just should invest time to setup some testing framework to get my answer? Do you manage any OpenBSD-based LTE + Wi-Fi access points? What's your experience? PS. Plese CC me in the replies, I don't have delivery enabled for misc. -- Regards, Mikolaj
