Poor WAN performance on BT HH5A <--> G.Fast MT992 Modem

From an SQM perspective, if your uploads kill the network performance not only for the uploading computer, but for all machines in your LAN, try the following:

config queue 'eth1'
        option ingress_ecn 'ECN'
        option itarget 'auto'
        option etarget 'auto'
        option verbosity '5'
        option qdisc 'cake'
        option script 'layer_cake.qos'
        option qdisc_advanced '1'
        option squash_dscp '0'
        option squash_ingress '0'
        option qdisc_really_really_advanced '1'
        option eqdisc_opts 'nat dual-srchost'
        option linklayer 'ethernet'
        option linklayer_advanced '1'
        option tcMTU '2047'
        option tcTSIZE '128'
        option linklayer_adaptation_mechanism 'default'
        option debug_logging '1'
        option iqdisc_opts 'nat dual-dsthost ingress'
        option interface 'wan'
        option tcMPU '96'
        option enabled '1'
        option overhead '44'
        option download '94500'
        option upload '7110'

The two tricks here are a) the dual-xxxhost keywords which instruct cake to share capacity fairly between internal IP addresses, that is your GDrive uploading computer will not be able to hog all/most upload capacity if other machines want to send something (but without competition the GDrive upload should still saturate your uplink). And ingress instructs cake to essentially automatically adjust the download shapers aggressiveness to the actual responsiveness of the incoming traffic. See https://openwrt.org/docs/guide-user/network/traffic-shaping/sqm-details for more information (also how to do a quick and dirty test whether the router's CPU might be overloaded).

Note with these shaper-rate settings you can at best expect the following TCP/IPv4 goodput (what speedtests measure and report):
9.4500 * ((1500-20-20)/(1500+44)) = 8.94 = Mbps
7.110 * ((1500-20-20)/(1500+44)) = 6.72 Mbps
as shaper rates are gross rates, while speedtests report net payload throughput...

I would, unless the router is already overloaded, set the shaper to 105000 and 8000 respectively, so those speeds that you get without SQM as speedtest results (and only decrease these if latency under load suffers too much).

Are you testing the ping from the same machine you are downloading from? Do you have a dedicated AP or does your router need to do both sqm and WiFi?

And do perform the MTR tests, Bill recommended... you can BTW run mtr from your router:
use opkg update ; opkg install mtr if mtr is not already installed and invoke it like:
mtr -ezb4w -c 120 pingbox1.thinkbroadband.com
for getting a nice report for ~2 minutes of pings (you can copy and paste the terminal output here in the forum), or use:
mtr -ezb4 pingbox1.thinkbroadband.com
to get a continuous updated mtr output in the terminal.
Side-note: pingbox1.thinkbroadband.com seems to be the host that Thinkbroadband uses for their network quality monitoring, so running mtr against that allows you to compare your results with those recorded and displayed with a thinkbraodbandprobe giving you a view from both sides.

1 Like