Augmented Traffic Control: A tool to simulate network conditions

Overview

Augmented Traffic Control

build-status-image pypi-version

Full documentation for the project is available at http://facebook.github.io/augmented-traffic-control/.

Overview

Augmented Traffic Control (ATC) is a tool to simulate network conditions. It allows controlling the connection that a device has to the internet. Developers can use ATC to test their application across varying network conditions, easily emulating high speed, mobile, and even severely impaired networks. Aspects of the connection that can be controlled include:

  • bandwidth
  • latency
  • packet loss
  • corrupted packets
  • packets ordering

In order to be able to shape the network traffic, ATC must be running on a device that routes the traffic and sees the real IP address of the device, like your network gateway for instance. This also allows any devices that route through ATC to be able to shape their traffic. Traffic can be shaped/unshaped using a web interface allowing any devices with a web browser to use ATC without the need for a client application.

ATC is made of multiple components that interact together:

  • atcd: The ATC daemon which is responsible for setting/unsetting traffic shaping. atcd exposes a Thrift interface to interact with it.
  • django-atc-api: A Django app based on Django Rest Framework that provides a RESTful interface to atcd.
  • django-atc-demo-ui: A Django app that provides a simple Web UI to use atc from a mobile phone.
  • django-atc-profile-storage: A Django app that can be used to save shaping profiles, making it easier to re-use them later without manually re-entering those settings.

By splitting ATC in sub-components, it make it easier to hack on it or build on top of it. While django-atc-demo-ui is shipped as part of ATC's main repository to allow people to be able to use ATC out of the box, by providing a REST API to atcd, it makes it relatively easy to interact with atcd via the command line and opens the path for the community to be able to build creative command line tools, web UI or mobile apps that interact with ATC.

ATC architecture

Requirements

Most requirements are handled automatically by pip, the packaging system used by ATC, and each ATC package may have different requirements and the README.md files of the respective packages should be checked for more details. Anyhow, some requirements apply to the overall codebase:

  • Python 2.7: Currently, ATC is only supported on python version 2.7.
  • Django 1.10: Currently, ATC is only supported using django version 1.10.

Installing ATC

The fact that ATC is splitted in multiple packages allows for multiple deployment scenarii. However, deploying all the packages on the same host is the simplest and most likely fitting most use cases.

To get more details on how to install/configure each packages, please refer to the packages' respective READMEs.

Packages

The easiest way to install ATC is by using pip.

pip install atc_thrift atcd django-atc-api django-atc-demo-ui django-atc-profile-storage

Django

Now that we have all the packages installed, we need to create a new Django project in which we will use our Django app.

django-admin startproject atcui
cd atcui

Now that we have our django project, we need to configure it to use our apps and we need to tell it how to route to our apps.

Open atcui/settings.py and enable the ATC apps by adding to INSTALLED_APPS:

INSTALLED_APPS = (
    ...
    # Django ATC API
    'rest_framework',
    'atc_api',
    # Django ATC Demo UI
    'bootstrap_themes',
    'django_static_jquery',
    'atc_demo_ui',
    # Django ATC Profile Storage
    'atc_profile_storage',
)

Now, open atcui/urls.py and enable routing to the ATC apps by adding the routes to urlpatterns:

...
...
from django.views.generic.base import RedirectView
from django.conf.urls import include

urlpatterns = [
    ...
    # Django ATC API
    url(r'^api/v1/', include('atc_api.urls')),
    # Django ATC Demo UI
    url(r'^atc_demo_ui/', include('atc_demo_ui.urls')),
    # Django ATC profile storage
    url(r'^api/v1/profiles/', include('atc_profile_storage.urls')),
    url(r'^$', RedirectView.as_view(url='/atc_demo_ui/', permanent=False)),
]

Finally, let's update the Django DB:

python manage.py migrate

Running ATC

All require packages should now be installed and configured. We now need to run the daemon and the UI interface. While we will run ATC straight from the command line in this example, you can refer to example sysvinit and upstart scripts.

atcd

atcd modifies network related settings and as such needs to run in privileged mode:

sudo atcd

Supposing eth0 is your interface to connect to the internet and eth1, your interface to connect to your lan, this should just work. If your setting is slightly different, use the command line arguments --atcd-wan and --atcd-lan to adapt to your configuration.

ATC UI

The UI on the other hand is a standard Django Web app and can be run as a normal user. Make sure you are in the directory that was created when you ran django-admin startproject atcui and run:

python manage.py runserver 0.0.0.0:8000

You should now be able to access the web UI at http://localhost:8000

ATC Code Structure

ATC source code is available under the atc directory, it is currently composed of:

  • atc_thrift the thrift interface's library
  • atcd the ATC daemon that runs on the router doing the traffic shaping
  • django-atc-api A django app that provides a RESTful interface to atcd
  • django-atc-demo-ui A django app that provides a simple demo UI leveraging the RESTful API
  • django-atc-profile-storage A django app that allows saving shaping profiles to DB allowing users to select their favorite profile from a list instead of re-entering all the profile details every time.

The chef directory contains 2 chef cookbooks:

  • atc A cookbook to deploy ATC. It also allows to deploy ATC in a Virtual Box VM in order to develop on ATC.
  • atclient Set up a Linux Desktop VM that can be used to test shaping end to end.

atcd

atcd is the daemon that runs on the router that does the shaping. Interaction with the daemon is done using thrift. The interface definition can be found in atc_thrift.thrift.

atc_thrift

atc_thrift defines the thrift interface to communicate with the atcd daemon.

django-atc-api

django-atc-api is a django app that provide a REST API to the atcd daemon. Web applications, command line tools can use the API in order to shape/unshape traffic.

django-atc-demo-ui

django-atc-demo-ui is a simple Web UI to enable/disable traffic shaping. The UI is mostly written in React

django-atc-profile-storage

django-atc-profile-storage allows saving profiles to DB. A typical use case will be to save a list of predefined/often used shaping settings that you want to be able to accessing in just a few clicks/taps.

Developing on ATC

To make ATC development easier, we use Virtual Box and Vagrant to provision and run a VM that will run the ATC daemon and the ATC UI from your git checkout.

Interacting with ATC will only shape the traffic within the VM and not on the host.

Setting up the environment

Note: vagrant is an easy way to set up a test environment, but virtualization will produce different results than a setup on bare-metal. We recommend using vagrant only for testing/development and using bare-metal for setups which require realistic shaping settings.

You will need to install VirtualBox, Vagrant and a couple of plugins:

  • VirtualBox
  • Vagrant
  • Chef DK
  • Install some vagrant plugins:
  • vagrant plugin install vagrant-berkshelf --plugin-version '>= 2.0.1'
  • vagrant plugin install vagrant-omnibus
  • Clone this repo: git clone [email protected]:facebook/augmented-traffic-control.git atc

Running ATC

Once in the repo, go to the chef/atc directory and run:

vagrant up trusty

This will take some time before it completes, once the VM is provision, SSH into it:

vagrant ssh trusty

You should now be able to access ATC at: http://localhost:8080/

Using the Sample Profiles

Once you've got ATC up and running, you can run the script utils/restore-profiles.sh to setup the set of default profiles.

The script needs to be passed a hostname:port with the location of your ATC instance:

utils/restore-profiles.sh localhost:8080

After doing this, you should see the 10 sample profiles listed below in your ATC instance:

  • 2G - Developing Rural
  • 2G - Developing Urban
  • 3G - Average
  • 3G - Good
  • Cable
  • DSL
  • Edge - Average
  • Edge - Good
  • Edge - Lossy
  • No Connectivity

Naturally, you cannot improve your natural network speed by selecting a faster profile than your service. For example, selecting the Cable profile will not make your network faster if your natural connection speed resembles DSL more closely.

Hacking on the code

Hacking on ATC is done from the host and tested in the VM. In order to reflect the changes, you will need to start the services manually.

Both atcd and atcui have their python libraries installed in a python virtualenv so you will need to activate the environment in order to be able to run the services.

The virtualenv is installed in /usr/local/atc/venv/bin/activate .

source /usr/local/atc/venv/bin/activate

Running the daemon

The atcd daemon is running under the root user privileges, all operations below needs to be done as root.

To run the daemon manually, first make sure it is not running in the background:

service atcd stop

And run the daemon:

atcd

Once you are happy with your changes and you want to test them, you will need to kill the daemon and restart it in order to apply the changes.

Running the API/UI

This is a django project and, when running the django built-in HTTP server, will detect code changes and reload automatically.

To run the HTTP REST API and UI:

cd /var/django && python manage.py runserver 0.0.0.0:8000
Owner
Meta Archive
These projects have been archived and are generally unsupported, but are still available to view and use
Meta Archive
Learning to Stylize Novel Views

Learning to Stylize Novel Views [Project] [Paper] Contact: Hsin-Ping Huang ([ema

34 Nov 27, 2022
Using deep learning model to detect breast cancer.

Breast-Cancer-Detection Breast cancer is the most frequent cancer among women, with around one in every 19 women at risk. The number of cases of breas

1 Feb 13, 2022
Winning solution of the Indoor Location & Navigation Kaggle competition

This repository contains the code to generate the winning solution of the Kaggle competition on indoor location and navigation organized by Microsoft

Tom Van de Wiele 62 Dec 28, 2022
shufflev2-yolov5:lighter, faster and easier to deploy

shufflev2-yolov5: lighter, faster and easier to deploy. Evolved from yolov5 and the size of model is only 1.7M (int8) and 3.3M (fp16). It can reach 10+ FPS on the Raspberry Pi 4B when the input size

pogg 1.5k Jan 05, 2023
Google AI Open Images - Object Detection Track: Open Solution

Google AI Open Images - Object Detection Track: Open Solution This is an open solution to the Google AI Open Images - Object Detection Track 😃 More c

minerva.ml 46 Jun 22, 2022
Deeper insights into graph convolutional networks for semi-supervised learning

deeper_insights_into_GCNs Deeper insights into graph convolutional networks for semi-supervised learning References data and utils.py come from Implem

Davidham3 17 Dec 16, 2022
On-device speech-to-intent engine powered by deep learning

Rhino Made in Vancouver, Canada by Picovoice Rhino is Picovoice's Speech-to-Intent engine. It directly infers intent from spoken commands within a giv

Picovoice 510 Dec 30, 2022
Code for the paper "Balancing Training for Multilingual Neural Machine Translation, ACL 2020"

Balancing Training for Multilingual Neural Machine Translation Implementation of the paper Balancing Training for Multilingual Neural Machine Translat

Xinyi Wang 21 May 18, 2022
SPT_LSA_ViT - Implementation for Visual Transformer for Small-size Datasets

Vision Transformer for Small-Size Datasets Seung Hoon Lee and Seunghyun Lee and Byung Cheol Song | Paper Inha University Abstract Recently, the Vision

Lee SeungHoon 87 Jan 01, 2023
CVNets: A library for training computer vision networks

CVNets: A library for training computer vision networks This repository contains the source code for training computer vision models. Specifically, it

Apple 1.1k Jan 03, 2023
GLM (General Language Model)

GLM GLM is a General Language Model pretrained with an autoregressive blank-filling objective and can be finetuned on various natural language underst

THUDM 421 Jan 04, 2023
This repository contains the code for the CVPR 2021 paper "GIRAFFE: Representing Scenes as Compositional Generative Neural Feature Fields"

GIRAFFE: Representing Scenes as Compositional Generative Neural Feature Fields Project Page | Paper | Supplementary | Video | Slides | Blog | Talk If

1.1k Dec 30, 2022
Cobalt Strike teamserver detection.

Cobalt-Strike-det Cobalt Strike teamserver detection. usage: cobaltstrike_verify.py [-l TARGETS] [-t THREADS] optional arguments: -h, --help show this

TimWhite 17 Sep 27, 2022
Pytorch implementation of SenFormer: Efficient Self-Ensemble Framework for Semantic Segmentation

SenFormer: Efficient Self-Ensemble Framework for Semantic Segmentation Efficient Self-Ensemble Framework for Semantic Segmentation by Walid Bousselham

61 Dec 26, 2022
Only works with the dashboard version / branch of jesse

Jesse optuna Only works with the dashboard version / branch of jesse. The config.yml should be self-explainatory. Installation # install from git pip

Markus K. 8 Dec 04, 2022
Multiple style transfer via variational autoencoder

ST-VAE Multiple style transfer via variational autoencoder By Zhi-Song Liu, Vicky Kalogeiton and Marie-Paule Cani This repo only provides simple testi

13 Oct 29, 2022
One Million Scenes for Autonomous Driving

ONCE Benchmark This is a reproduced benchmark for 3D object detection on the ONCE (One Million Scenes) dataset. The code is mainly based on OpenPCDet.

148 Dec 28, 2022
Anchor-free Oriented Proposal Generator for Object Detection

Anchor-free Oriented Proposal Generator for Object Detection Gong Cheng, Jiabao Wang, Ke Li, Xingxing Xie, Chunbo Lang, Yanqing Yao, Junwei Han, Intro

jbwang1997 56 Nov 15, 2022
Official repository for Jia, Raghunathan, Göksel, and Liang, "Certified Robustness to Adversarial Word Substitutions" (EMNLP 2019)

Certified Robustness to Adversarial Word Substitutions This is the official GitHub repository for the following paper: Certified Robustness to Adversa

Robin Jia 38 Oct 16, 2022
PyTorch implementation of image classification models for CIFAR-10/CIFAR-100/MNIST/FashionMNIST/Kuzushiji-MNIST/ImageNet

PyTorch Image Classification Following papers are implemented using PyTorch. ResNet (1512.03385) ResNet-preact (1603.05027) WRN (1605.07146) DenseNet

1.2k Jan 04, 2023