Once on the access server, Ubuntu Server 16.04 and Accel-ppp noticed the following messages in the /var/log/kern.log file:
kernel: [365970.550498] HTB: quantum of class 10001 is big. Consider r2q change. kernel: [365970.550547] HTB: quantum of class 10A49 is big. Consider r2q change. kernel: [365979.545580] HTB: quantum of class 10001 is big. Consider r2q change. kernel: [365979.545621] HTB: quantum of class 10BD6 is big. Consider r2q change. kernel: [365995.601973] HTB: quantum of class 10001 is big. Consider r2q change. kernel: [365995.602031] HTB: quantum of class 11705 is big. Consider r2q change.
First I tried to track which interfaces are being raised at this moment:
tail -f /var/log/kern.log | grep "quantum of class 10001 is big" tail -f /var/log/accel-ppp/accel-ppp.log | grep "create interface"
But as it turned out, in the configuration of /etc/accel-ppp.conf, in the shaper section, quantum was not specified, and if it is not specified, it is calculated.
So I opened the Accel-ppp configuration file:
sudo nano /etc/accel-ppp.conf
And pointed it out:
[shaper] quantum=1500 #moderate-quantum=1
Also changed the up-limiter from “htb” to “police” for better performance, and also to get the speed cut off correctly with a large number of sessions (at 3000+ ipoe sessions with htb problems started, the sending rate fell to 1 MB/sec):
#up-limiter=htb up-limiter=police
To apply the change for new sessions without terminating the active ones, we will force Accel-ppp to re-read the configuration file:
telnet 127.0.0.1 2000 reload
Or we will reload Accel-ppp, but all sessions will be terminated:
sudo /etc/init.d/accel-ppp restart
After that, the error disappeared.
See also my article – Accel-ppp installation