CS244 PA3: Verus vs. Sprout

Andrew Lim ([email protected]) and Ryan Hermstein ([email protected])

Code structure:

• The Sprout code is located in alfalfa/src/examples. We use the sendonly branch of alfalfa.
• The Verus code is located in verus/src.
• The scripts that run our experiments as well as our variable trace generator script are in scripts.
• Our modified plotting scripts for Verus and the combined Verus-Sprout throughput graph are in plot_scripts.
• Our implementation of RED for Mahimahi is located in mahimahi/src/packet/red_packet_queue.hh.

Setting up an EC2 Instance

If you are unfamiliar with EC2, consider looking at CS244's guide to getting started and Amazon's beginner guide.

To run the experiments, you will need to set up an EC2 instance from our public AMI. Make sure that your region is set to US West (N. California) region, as this is where the AMI is registered. You can change your region in the top right corner of the upper navigation bar. In the EC2 Management Console, click on AMIs under Images in the left sidebar. Make sure to search for "Public Images" instead of "Owned by Me". Our AMI id is: ami-38afd458 and the AMI name is: cs244-pa3-verus-sprout-alim16-rherms. Right click the AMI and click "Launch" to launch an instance.

The experiments should be run on a single m4.xlarge instance with 30 GiB of General Purpose SSD storage (this will be the default when you configure instance settings). When configuring the security group, you will want to add an All UDP rule with port range 0 - 65535 and source set to My IP. Once you have configured these settings, the instance is ready to be launched.

Running the Experiments

After launching an EC2 instance from our public AMI, SSH into your instance. All the executables should be already compiled. Thus, you can run the experiments immediately using the following commands:

cd PA3
python run_experiment.py

The experiments will take ~30 minutes to run. The resulting graphs will be placed in the results subdirectory within the PA3 directory. SCP the graph files to your local machine, and open them in a web browser to view. The plot html file names will include the following information: verus or sprout (which congestion control algorithm), var or alma (indicate whether the trace used was the synthetic variable trace or the Alma Street trace), and throughput or delay (what kind of plot). The plot that compares the throughputs is named "var_throughput_comp.html".

Don't forget to stop your instance after obtaining the result graphs! Also, if you want to close the SSH connection to your instance while the experiments are running, look into "screen".

Note: If you would like to re-run the experiments, you should first run the cleanup script to delete any logs/outputs from the previous run. From within PA3:

./cleanup_results.sh

Also, if you would like to run the experiments using the modified Verus parameters we considered, you will have to recompile the Verus server and client. From within PA3:

cd verus/src
cp verus_modified.hpp verus.hpp
make clean && make

You can then navigate back up to the topmost level of the PA3 directory and run the run_experiment.py script. Likewise if you would like to return to the original Verus params, from within PA3:

cd verus/src
cp verus_original.hpp verus.hpp
make clean && make

If you would like to run the experiments including some of the additional traces (ATT and Verizon), use the run_experiment_full.py script.