EFM Testbed

Directory Structure

├── configurations -- Measurement configurations
├── go_analysis -- Post-facto analysis scripts written in Go (Observer Logic)
├── python_analysis -- Testbed orchestration software
└── traffic -- Quic client/server code for traffic generation

Publication

This work has been created in the context of the following publication:

• Ike Kunze, Klaus Wehrle, and Jan Rüth: L, Q, R, and T - Which Spin Bit Cousin Is Here to Stay?. In ANRW '21: Proceedings of the Applied Networking Research Workshop

If you use any portion of our work, please consider citing our publication.

    @Inproceedings{2021-kunze-efm-evaluation,
    author = {Kunze, Ike and Wehrle, Klaus and R{\"u}th, Jan},
    title = {L, Q, R, and T - Which Spin Bit Cousin Is Here to Stay?},
    booktitle = {ANRW '21: Proceedings of the Applied Networking Research Workshop},
    year = {2021},
    month = {July},
    doi = {10.1145/3472305.3472319}
    }

General Remarks

This testbed is designed for use on a Linux system and has only been tested for Ubuntu 18.04.

Installation

• Install aioquic (see: https://github.com/COMSYS/aioquic)

  • sudo apt install libssl-dev python3-dev
  • git clone git@github.com:COMSYS/aioquic.git / git clone https://github.com/COMSYS/aioquic.git
  • cd aioquic/ && sudo pip install -e . (make sure that pip links to pip3)

• Further setup for the EFM measurement framework

  • sudo apt install mininet
  • Install BCC (https://github.com/iovisor/bcc)
    • sudo apt-get install bpfcc-tools linux-headers-$(uname -r)
  • NOTE: monitor_queue_bpf_enqueue_only.py has been in use with python2

• Install analysis scripts

  • Install libpcap: sudo apt-get install libpcap-dev
  • Install golang: sudo snap install go
  • Build the different analysis scripts
    • cd /path/to/efm-evaluation-anrw/go_analysis/ && go build am-pcap-analyzer.go && cd -
    • cd /path/to/efm-evaluation-anrw/go_analysis/ && go build analyse_queueMonitor.go && cd -
    • cd /path/to/efm-evaluation-anrw/go_analysis/ && go build queueMonitor_burstsize_calculator.go && cd -
      • This is an additional tool that can be used to determine the burst sizes

• Files for download volumes

  • If you plan to use the H3 mode, please place the corresponding files in traffic/htdocs/
  • Name them like 50k!lossrandom-1.file where the first part is the experiment description in the configuration file

Quick Start

• Run a measurement

  • After these steps, you should be able to run a simple example using
  • cd /path/to/framework/python_analysis/ && python3 simulator.py --config paper_demo

• Run an analysis

  • python3 perform_analysis.py --path data/paper_demo/results/
  • Note: this requires that the go-based analysis scripts have been built

Experiment Workflow

1. Create an experiment EXPERIMENT_ID.json configuration file in configurations
   • The analysis framework is currently designed to support analyses for loss occurring on link s3-eth1
2. Start the experiment using python3 simulator.py --config EXPERIMENT_ID.json
3. The analysis is then performed afterwards
   • python3 perform_analysis.py --path /path/to/measurements/

Content Details

configurations

There are three configurations used for the evaluation (paper_eval_(random|gemodel|congestion).json) as well as one demo configuration (paper_demo.json) that can be used to test the setup. The configurations contain the following parameters:

• comment: Short description of the overall configuration
• experiment: specification of the actual traffic settings
  • iterations: defines how often each of the given flow_tests will be performed
  • flow_tests: list of experiments to be performed
    • (src|dst)_(host|port): define src/dst hosts/ports. Note that dst_port 1234 is currently hardcoded in the tc-based network setup scripts.
    • description: short id of the experiement that will be prefixed to the measurement output
    • link_configs: configurations of the involved links. "link" specifies the interface, "netem_args" the corresponding netem arguments.
    • (synthetic|http)_traffic: defines the used traffic type
      • http_traffic - file_size: Defines the file that will be downloaded. Make sure that a corresponding file named "{file_size_argument}.file" is stored in traffic/htdocs
      • synthetic_traffic - packets: Define the number of packets that will be transmitted
      • synthetic_traffic - duration: Define the transmission duration
    • measurement_techniques: enable/disable the desired measurement techniques
• hosts: define the involved end-hosts as well as their ip addresses

traffic

There are two client/server pairs.

• client.py / server.py: This implementation uses the datagram mode to continously transmit symmetric traffic between client and server
• http3_client.py / http3_server.py: This implementation is a 'standard' http3 connection. It extends the http3 example of aioquic with corresponding calls to initialize the used EFM variants
  • The ssl_key.pem and ssl_cert.pem we use are also taken from aioquic.

python_analysis

This directory contains the actual measurement infrastructure

• simulator.py: main script for performing the measurements
• perform_analysis.py: script for analyzing the measurement results
  • analyzer_loss.py: helper file for the analysis
• custom_mininet_topo.py: specifies the underlying mininet topology
• monitor_queue_bpf_enqueue_only.py: bpf script used to observe the queue state
• average_burst_size_calculator.py: can be used to analyze the observed burst sizes
• [add|remove]_network_namespace.sh: scripts to add helper network namespaces for easier access to the virtual hosts

go_analysis

This directory contains the observer logic implemented in go

• am-pcap-analyzer.go: Observer logic for the EFM techniques focussing on loss
• analyse_queueMonitor.go: Helper analyzer used to derive the groundtruth
• queueMonitor_burstsize_calculator.go: additional tool that can be used to determine the burst sizes