CN contact: Maximilian Bachl

Cocoa is a qdisc which maximizes throughput for each flow while keeping the buffer minimal. For a more detailed description check out the upcoming paper.

To compile the kernel module run

make

to install it and load it into the kernel run

sudo make install

Next, change into the iproute2 folder cd iproute2 and run

make
sudo make install

To make the configuration for the module visible to the tc utility we have to set

export TC_LIB_DIR=<path to the repository>/cocoa-qdisc/iproute2/tc

Finally, you can use the qdisc on an interface:

sudo -E tc qdisc replace dev <interface> root cocoa

tc also allows you to specify options like this:

sudo -E tc qdisc replace dev <interface> root cocoa initial_quantum 3028 quantum 3028

To run experiments, make sure you have py-virtnet (GitHub repository) installed and then run

sudo bash -c 'echo > /sys/kernel/debug/tracing/trace' && sudo bash -c 'echo 1 > /sys/kernel/debug/tracing/tracing_on' && sudo python3 test.py --path_to_tc_module <path to the repository>/cocoa-qdisc/iproute2/tc --rate 20 --delay_to_add 1 --time 30 --qdisc cocoa --change 1 --cc cubic

After running experiments with cocoa, you can look at detailed output in the kernel tracing file at /sys/kernel/debug/tracing/trace.

To create plots of a run and show further statistics, first compile wintracker:

go build -o wintracker wintracker.go

Then, if you have a file called sender_fq_codel_cubic_1_20_120_1.0_bw_1571822805075.pcap in the pcaps directory you can run the plotting script:

./plot_rtt_and_bandwidth.py sender_fq_codel_cubic_1_20_120_1.0_bw_1571822805075.pcap

All tests were performed on kernel 4.19.0-6-amd64 on Debian Buster. We use Python 3.7.2. Our go version is go1.10.2 linux/amd64.