Intro

This repository contains logic for reproducing results from the ABC congestion control research paper by Goyal et al.

Logic for ABC is in the file - mahimahi/src/packet/cellular_packet_queue.cc

---

Setup

1. Clone the repo: https://github.com/sktollman/ABC-1.git.
2. Create a Python 2 virtual environment within ABC-1 and activate it.
3. Run the setup script: sh setup/setup.sh

The setup script will install dependencies (e.g. pip packages, debian packages, the Stanford Pantheon for congestion control), install mahimahi, and prepare your Kernel for running experiments.

Note that this script will modify the allowed congestion control modules in the kernel, so see below for manually installing congestion control modules if you would prefer not to run the script.

Installing Congestion Control Modules

We need cubic, vegas, and bbr. If you are running on a Google Cloud Compute instance, this part is already taken care for you if you followed the steps in the previous section.

If you are not running on GCC

To see which modules are loaded in your kernel, run:

$ /sbin/sysctl net.ipv4.tcp_available_congestion_control

To see which can be run without root permissions, run:

$ /sbin/sysctl net.ipv4.tcp_allowed_congestion_control

To see the available modules, run

$ ls -la /lib/modules/$(uname -r)/kernel/net/ipv4

To load modules, run modprobe. (eg: sudo modprobe -a tcp_vegas). You should now see these modules in the list of available congestion control protocols. You may still need to enable some to run without root permissions. In this case, you will need to append to that file. When writing to the file you want to make sure that you don't remove existing protocols. (e.g. $ echo cubic reno bbr vegas | sudo tee /proc/sys/net/ipv4/tcp_allowed_congestion_control)

Reproduction

To generate the graphs in our report, run:

$ sh run_all.sh
$ sh generate_plots.sh

When this completes, you will have plots for Figure 1, Figure 2a, Figure 2b, and the variable ack path experiment saved in a /plots/ directory.

Figure 2

experiments.py contains the code to reproduce Figures 1 and 2. This runs traces against each protocol and optionally saves the results to a csv file. The --schemes command-line parameter takes in a comma-separated list of protocols to run. If the option is not given, results for all protocols are generated. The --experiment parameter takes a string in figure1, figure2a, figure2b, bothlinks, pa1 according to which experiment we are running.

For example, to generate ABC and Verus results for Figure 2a, run python experiment.py --experiment figure2a --schemes abc verus.

The --reuse-results and --run-full parameters allow the developer to specify exactly which experiments to re-run, and which to only display results from given that a results file exists. For example, python experiment.py --experiment figure2a --schemes abc sprout --reuse-results abc will return Figure 2a results for ABC and Sprout, but will only fully re-run the experiment for Sprout, because ABC is skipped. Not specifying any of these commands will spur a fresh run of all the protocols given in --schemes.

Lastly, the command-line parameter --csv-out allows you to specify a filename to which experiment results will be printed in csv format. This file is consumed by the plotting scripts described in the next section.

For Figure 1 specifically, the --traces and --reuse-results-fig1 options are available. The space separated list for --reuse-results-fig1 takes the format [protocol]:[trace]

Generating Figure Plots

figure1_plot.py [data-filename] [plot-filename] takes a csv results file from experiments.py and creates a matplotlib graph with the same format as the figure from the original paper.

figure2_plot.py [data-filename] [plot-filename] takes a csv results file from experiments.py and creates a matplotlib graph with the same format as the figure from the original paper.

Optional arguments:

• --original-figure plots the provided data in comparison to the results from the original ABC paper.
• --cloud plots many datapoints as a point cloud to show variation across runs.
• --limit limits to only displaying protocols from the original paper
• --better-box adds a 'better' arrow indicating how to read the graph

Example:

$ python experiment.py --schemes all --experiment figure2a --csv-out results.csv
$ python figure2_plot.py results.csv plot.svg -o 2a -c -l -b