On Fri, Apr 24, 2020 at 9:32 AM Dave Taht <[email protected]> wrote: > > It looks to me you should reduce your download settings a bit for the > docsis service. > > On Fri, Apr 24, 2020 at 9:26 AM Kenneth Porter <[email protected]> wrote: > > > > Here's what the fiber connection looks like with no SQM applied: > > > > <http://www.dslreports.com/speedtest/62767361> > > That's miserable. 480ms latency on fiber?? You can do so much better. > But why centos? sure the sqm-scripts work with that but you should be > able to shape 45Mbits with even a wndr3800. openwrt works great on x86 > hw, also. :) > > do you get dedicated ipv6 with that AT&T service? > > What will be the vpn type? ipsec, terminating on the router, works > well with fq_codel because the hash is propagated to the tunnel, > wireguard and openvpn currently do not.
And cake does not propagate the hash because it recalculates it. So if your use case is ipsec, terminating on the router, and it's primarily vpn traffic, I would use fq_codel, rather than cake. Any other use case, cake. :) These days I slam another cake instance in front of my wireguard tunnels, set to a slightly lower bandwidth than the next hop (20% less). I'm always willing to live with a bandwidth hit in order to generally get better latency on my vpn'd traffic. YMMV. > I have been tempted, with smb3 supporting crypto e2e, and ipv6, to > ditch the vpn entirely. But only tempted! > > > _______________________________________________ > > Bloat mailing list > > [email protected] > > https://lists.bufferbloat.net/listinfo/bloat > > > > -- > Make Music, Not War > > Dave Täht > CTO, TekLibre, LLC > http://www.teklibre.com > Tel: 1-831-435-0729 -- Make Music, Not War Dave Täht CTO, TekLibre, LLC http://www.teklibre.com Tel: 1-831-435-0729 _______________________________________________ Bloat mailing list [email protected] https://lists.bufferbloat.net/listinfo/bloat
