sqm-scripts-nss

Smart Queue Management Scripts for OpenWRT for use with NSS optimized builds.

NSS FQ-Codel proves very effective at maintaining low latency under load, while causing minimal CPU load on the router.

Currently only supports nssfq-codel and no traffic classification / marking due to limitations of the current driver.

Requirements

• An SQM enabled OpenWRT build such as https://github.com/ACwifidude/openwrt
• sqm-scripts package
• luci-app-sqm package (for configuration from the GUI)

Installation

Manual Installation

• Just copy the nss-rk.qos and nss-rk.qos.help files to /usr/lib/sqm on your router

Package Installation

• Download the .ipk package file from the releases page
• Go to the System -> Software menu on your router and upload the .ipk package file

Installation via feeds

If you're building OpenWRT yourself, you can add this script to your build with a feed:

echo "src-git sqm_scripts_nss https://github.com/rickkdotnet/sqm-scripts-nss.git" >> feeds.conf
./scripts/feeds update
./scripts/feeds install sqm-scripts-nss

Now you can find the script in menuconfig under 'Extra packages'.

Configuration

• Go to Network -> SQM QoS in luci
• Add a a queue or change your existing one
• Select your physical uplink interface (usually eth0)
• Sheck the 'enable this SQM instance' checkbox
• Enter your down and upload speeds, 95% of your actual line speed is a good ballpark figure
• Go to the queue discipline tab and select "fq_codel" as discipline and and "nss-rk.qos" for the script
• Configure other parameters if you want, although the defaults should work well. If you like to play with the codel interval, you can do by entering 'interval XXms' in the 'advanced option string'.
• Click "save and apply"

If all went well you should now be able to enjoy lower latency under load, with minimal CPU load on the router.

If it's working the output from tc should look something like this:

root@nighthawk:~# tc -s qdisc show dev eth0
qdisc nsstbl 1: root refcnt 2 buffer/maxburst 4500b rate 36Mbit mtu 1514b accel_mode 0
 Sent 2692502 bytes 7109 pkt (dropped 0, overlimits 190 requeues 0)
 backlog 0b 0p requeues 0
qdisc nssfq_codel 10: parent 1: target 5ms limit 1001p interval 50ms flows 1024 quantum 300 set_default accel_mode 0
 Sent 2692502 bytes 7109 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
 maxpacket 1518 drop_overlimit 0 new_flow_count 4024 ecn_mark 0
 new_flows_len 0 old_flows_len 5

root@nighthawk:~# tc -s qdisc show dev nssifb
qdisc nsstbl 1: root refcnt 2 buffer/maxburst 45000b rate 360Mbit mtu 1514b accel_mode 0
 Sent 2182202 bytes 8391 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
qdisc nssfq_codel 10: parent 1: target 5ms limit 1001p interval 50ms flows 1024 quantum 1514 set_default accel_mode 0
 Sent 2182202 bytes 8391 pkt (dropped 0, overlimits 0 requeues 0)
 backlog 0b 0p requeues 0
 maxpacket 1518 drop_overlimit 0 new_flow_count 5081 ecn_mark 0
 new_flows_len 0 old_flows_len 1

Configuration notes

• DO try lowering the codel interval to the worst-case latency of the internet services you frequently use. This make codel more aggressive/quicker to react. I use "interval 50ms" in the advanced options string
• DO try lowering the codel target a little if you're on a low latency connection, it could shave off another ms or so.
• DO try adding "quantum 304" to the advanced options string for the lowest latency for games, voip etc. This essentially gives <304 bytes packets a 5x higher priority than 1518 packets to get dequeued.
• DON'T set the queue size so small that you have taildropping (drop_overlimit counter in tc -s increasing), as this causes significant bufferbloat in the first few seconds of a speedtest (try pinging with a short interval while starting your test). Popular opinion is that 1000 packets should be enough for gigabit, but I needed more than 2000 packets on my 350 Mb/s nssifb interface. This might be a quirk of NSS/nssifb.
• DO your own testing and draw up your own conclusions. First there's a lot of hear-say on the internet, and second your situation might very well be different.

FYI, as of March 2023, my tc config looks like this:

qdisc nsstbl 1: dev eth0 root refcnt 2 buffer/maxburst 4554b rate 37Mbit mtu 1518b accel_mode 0
qdisc nssfq_codel 10: dev eth0 parent 1: target 3.5ms limit 304p interval 50ms   flows 1024 quantum 304 set_default accel_mode 0```
qdisc nsstbl 1: dev nssifb root refcnt 2 buffer/maxburst 45540b rate 360Mbit mtu 1518b accel_mode 0
qdisc nssfq_codel 10: dev nssifb parent 1: target 3.5ms limit 2964p interval 50ms flows 1024 quantum 304 set_default accel_mode 0

Known bugs, limitations

• Due to limitations of the driver:
  • Only fq-codel is supported as a queue discipline
  • No marking or traffic classification is currently possible, this also means that DSCP squashing does not work
  • ECN marking is not supported
• The script does not does anything with the Link Layer Adaptation fields.
• On kernel 5.10 the script does not remove the nssifb interface if it's stopped, because removing or even bringing down the interface frequently crashed my router. This could cause problems if you're switching to a script which set up regular ifb4ethX interfaces. You probably need to reboot if you want to switch to another SQM script.
• On kernel 5.15 the above problem has been resolved, but in some cases (I think under high load), the router crashes on ip link up of nssifb, especially when called from hotplug. Because of this, there's workaround in the script that prevents the script from executing when it's called from hotplug.