Interaction between Cake-QOS and OpenVPN

[dave14305]

Update: ah, I suppose VPN is not up at this point? How should I address that?

Find some hook to restart QoS once the VPN client is up. Outside of my expertise.

 

[Lynx]

Find some hook to restart QoS once the VPN client is up. Outside of my expertise.

Thanks - got it based on the wiki:

admin@RT-AX86U-4168:/tmp/home/root# cat /jffs/scripts/openvpn-event
#!/bin/sh
service restart_qos

All works well now it seems.

By the way, did my calculation of WAN packet overhead of 53 as determined above look right to you?

Do you not think it would be fair to classify this experiment as a success? My bufferbloat seems to be well handled now, even up to 40/20 on my LTE 4G connection with VPN active:

Bufferbloat and Internet Speed Test - Waveform

View the full results, and test your own bufferbloat

www.waveform.com

Speed test result

I tested my connection at DSLReports.com/speedtest ..

www.dslreports.com

I hope this thread might be helpful for anyone else in a similar boat to me.

Regarding the bandwidth sacrifice, sacrificing the 10-20 / 5 bandwidth in exchange for this stability seems OK. Perhaps I could squeeze yet more out of this though.

On that point, has anyone had much success with 'autorate-ingress' on an LTE connection I wonder?

 

[dave14305]

Do you not think it would be fair to classify this experiment as a success?

Doesn’t matter what I think. You’re the judge. If it works for your specific situation, that’s a win.

 

[ttgapers]

Find some hook to restart QoS once the VPN client is up. Outside of my expertise.

You can add the hook for the restart in the vpn-client-connect call. I currently use it for email notifications when clients connections are made.

see: https://github.com/RMerl/asuswrt-merlin.ng/wiki/User-scripts

 

[Lynx]

Thanks all.

Am I correct in thinking the overhead as calculated above should be applied to both 'tun11' and 'br0'? I checked and the packet lengths are consistent when generated from ping commands from router and from entities connected to router. From router I get difference (137-84=53); from entities connected to router I get difference (113-60=53).

Should 'nat' be applied on 'tun11' and 'nonat' applied on 'br0'?

So to sum up, does:

qdisc cake 8014: dev br0 root refcnt 2 bandwidth 40960Kbit besteffort dual-dsthost nonat wash ingress no-ack-filter split-gso rtt 100ms noatm overhead 53
qdisc cake 8013: dev tun11 root refcnt 2 bandwidth 20480Kbit diffserv3 dual-srchost nat nowash no-ack-filter split-gso rtt 100ms noatm overhead 53

look right?

Finally, ought not really an ifb4tun11 interface to be created? That seems to be how it is done in SQM on OpenWRT. Or is there some benefit in working on br0? I notice that ping commands issues on router do not seem to appear on br0.