Linux – Resetting a queue discipline back to the default pfifo_fast one

Nice Script, I like the part with the time depended dynamic download speed. ;) Anyway, I am doing excactly what you want. I did it with htb, but switched a few months ago to hfsc. The trick is to limit q3 with rate to a very low rate, but give it with ceil the full bandwith. q1 on the other hand gets a higher rate and the same ceil. One realworld example:

tc class add dev ppp0 parent 1: classid 1:1 htb rate 100kbps ceil 100kbps
tc class add dev ppp0 parent 1:1 classid 1:10 htb rate 60kbps ceil 100kbps prio 1 
tc class add dev ppp0 parent 1:1 classid 1:11 htb rate 39kbps ceil 100kbps prio 2 
tc class add dev ppp0 parent 1:1 classid 1:12 htb rate 1kbps ceil 100kbps prio 3 

Q1 would be 1:10

Q2 would be 1:11

Q3 would be 1:12

Maybe your problem lies within your sophisticated script. Try a more lightweight version for debbuging.

tc -s -d class show dev <interface> should show some stats about the queues. Let´s show us the output if q1 has nothing to do and q3 doesn´t use the full bandwith. Also you could give hfsc a try.

One Remark: QOS or Traffic Control on your incoming traffic is pretty useless in most configurations. Typically you are in a reciever role who has no power over recieving what amount of data/time. You could use Ingress-Queueing but it won´t help much, it only discardes already recieved packets.

That is not as bad as it sounds because most dial-in connections are nowadays asymetric, which means that the transmitting direction is the bottleneck. (Always from your point of view)

ps -A -o pid,state,command | awk '{ if ($2 == "R") print }'
 1605 R /usr/bin/skype
30655 R ps -A -o pid,state,command

or use htop and sort(F6) by "S"

or to troubleshoot highload values:

while [ 1 ] ; do ps -A -o pid,state,command | \
   awk '{ if ($2 != "S") print }' | \
   grep -v "ps -A -o pid,state,command" ; sleep 3 ; done