- Follow steps in https://github.com/google/bbr/blob/master/Documentation/bbr-quick-start.md
- If you don't have a Google compute engine, follow the following steps:
- Install lilo https://linux.die.net/man/8/lilo
- Obtain kernel source with BBR
git clone git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next.git
cd net-next/
make olddefconfig
- Make sure CONFIG_BBR is set to true and FQ modules are enabled. Here is the .config file I used to successfully install the kernel.
- Install kernel
- change /etc/lilo.conf to include the installed kernal
make -j 8 && make -j 8 modules_install && make install
change /etc/lilo.conf to include the installed kernal
/sbin/lilo
- reboot
- confirm fq qdisc is installed and that BBR is avaiable:
tc qdisc show
cat /proc/sys/net/ipv4/tcp_available_congestion_control
Follow steps in http://mahimahi.mit.edu/
wget http://downloads.es.net/pub/iperf/iperf-3-current.tar.gz
- Install python 2.7
- Install python matplotlib http://matplotlib.org/
A simple dumebell topology is used where two nodes connected with a 1Gbits/sec link. Each node is an off-the-shelf PC configured with 4 (8 hyper-threaded) Core Intel i7-3770 3.40 GHz CPU with 16GB of memory.
Change the server Ip adderss to be your server Ip address.
Run following commands and view results in /tmp/$cc-maxb-1G-basertt-0-BG-0-tcpdump-N-interval-.1-exp1/
for cc in "cubic" "bbr"
do
bash exp_base_performance.sh $cc '1G' 0 0 'N' '.1' 'exp1'
done
Run following commands and view results in /tmp/$cc-maxb-1G-basertt-0-BG-0-tcpdump-N-interval-.1-exp2/
for cc in "cubic" "bbr"
do
bash exp_base_performance.sh $cc '1G' 0 0 'N' '.1' 'exp2'
done
Run following commands and view results in /tmp/$cc-maxb-1G-basertt-0-BG-200M-tcpdump-N-interval-.1-$exp/
for exp in "exp1" "exp2"
do
for cc in "cubic" "bbr"
do
bash exp_base_performance.sh $cc '1G' 0 '200M' 'N' '.1' $exp
done
done
Run following commands and view results in /tmp/mix-maxb-1G-basertt-0-BG-0-tcpdump-N-interval-.1-mix/
bash exp_base_performance.sh 'mix' '1G' 0 '' 'N' '.1' 'mix'