Google has come up with a new way of handling network congestion, by pretending there is no congestion. OK, maybe exaggerated, but who can blame me when they outrightly state that "1980s-era algorithm assumes that packet loss means network congestion".
BBR stands for Bottleneck Bandwidth and Round-trip propagation time. According to Google, YouTube network throughput improved by 4 percent on average globally. Wordpress also experience massive improvements of up to 2700 times better than the previous best loss-based congestion control.
Sounds impressive? Why not let's give it a try.
Back up your server before proceeding. You're likely going to need to change Linux Kernal and other system level changes.
Check the Linux Kernel Version
uname -mrs
If it's any version less than 4.9, you need to upgrade it. Otherwise, you can skip ahead to the next section.
Search for the latest linux kernal in the repository.
apt search linux-generic
At the time of writing, the latest version is linux-image-4.11.0-14-generic. TCP BBR is only introduced in 4.9 and higher, so choose any version 4.9 or higher and install.
apt install linux-image-4.11.0-14-generic
You need to reboot the server after this.
reboot
Enable it at sysctl.conf
nano /etc/sysctl.conf
Append the following lines
net.core.default_qdisc=fq
net.ipv4.tcp_congestion_control=bbr
Apply the settings
sysctl --system
Verify the new settings
sysctl net.core.default_qdisc
sysctl net.ipv4.tcp_congestion_control
You should see the output as what you typed in the sysctl.conf
Does it work? Maybe. When I did a speed test before and after implementing TCP BBR, there was hardly any noticable improvement in speedtest.