BurstRadar

This repository contains implementation of BurstRadar system using P4 V1Model Architecture on simple_switch target. BurstRadar system is presented in the paper: BurstRadar: Practical Real-time Microburst Monitoring for Datacenter Networks by Raj Joshi et. al., published in the Proceedings of the ACM 9th Asia-Pacific Workshop on Systems (APSys’18).

BurstRadar detects a microburst in the dataplane, captures a snapshot of telemetry information of all the involved packets, and further exports this telemetry information to a monitoring server in an out-of-band manner. A detailed explanation and background of BurstRadar system is provided in the aforementioned paper.

Steps to test BurstRadar system

A. Environment Setup

1. Install Oracle VirtualBox.
2. Download the VM Image (P4 Tutorial 2019-08-15).
3. Import P4 Tutorial 2019-08-15.ova appliance in VirtualBox.
4. Start the VM in VirtualBox and execute the following:

   • Change to /home/vagrant directory.

     vagrant@p4:~$ cd /home/vagrant
     

   • Clone the p4lang/tutorials repository.

     vagrant@p4:~$ git clone https://github.com/p4lang/tutorials.git
     

   • Uninstall python-scapy and its dependent packages.

     vagrant@p4:~$ sudo apt-get remove --auto-remove python-scapy
     

   • Download and install Scapy 2.4.3.

     vagrant@p4:~$ git clone https://github.com/secdev/scapy.git 
     vagrant@p4:~$ cd scapy
     vagrant@p4:~/scapy$ sudo python setup.py install
     

   • Set environment PATH to scapy directory.

     vagrant@p4:~/scapy$ gedit ~/.bashrc
     

     • Add the following line to .bashrc file, save and exit.

       export PATH="/home/vagrant/scapy:$PATH" 
       

     • Source .bashrc file.

       vagrant@p4:~/scapy$ source ~/.bashrc
       

   • Install tcpreplay package which is needed for executing sendpfast() scapy function.

     vagrant@p4:~$ sudo apt-get install tcpreplay
     

   • Change to the exercises directory.

     vagrant@p4:~/scapy$ cd ../tutorials/exercises/
     

   • Clone the burstradar repository and move to that directory.

     vagrant@p4:~/tutorials/exercises$ git clone https://github.com/harshgondaliya/burstradar.git
     vagrant@p4:~/tutorials/exercises$ cd burstradar
     vagrant@p4:~/tutorials/exercises/burstradar$
     
     

B. Running BurstRadar

1. In the /home/vargrant/tutorials/exercises/burstradar/ directory, execute:

   vagrant@p4:~/tutorials/exercises/burstradar$ sudo make run
   

   BMv2 Mininet CLI starts.
2. Open a new terminal and execute the following:
   • Start CLI

     vagrant@p4:~$ simple_switch_CLI
     

     Connection to the BMv2 simple_switch through thrift-port is started.
   • Set default values of bytesRemaining and index registers

     vagrant@p4:~$ simple_switch_CLI
     Obtaining JSON from switch...
     Done
     Control utility for runtime P4 table manipulation
     RuntimeCmd: register_write bytesRemaining 0 0
     RuntimeCmd: register_write index 0 0
     

   • Set mirror port for a given session id (In our case, session id = 11)

     RuntimeCmd: mirroring_add 11 4
     

3. In BMv2 Mininet CLI, execute:

   mininet> xterm h1 h2 h3 h4 h3
   

   Note: Two xterm displays for h3 are started.
   • In h4's xterm display, execute:

     ./receive.py
     

   • In the first xterm display of h3, execute:

     ./receive.py
     

   • In the second xterm display of h3, execute:

     iperf -s -w 2m
     

   • In h2's xterm display, execute:

     iperf -c 10.0.3.3 -w 2m -t 35
     

     This ensures that approx. 4-4.5 Mbps background traffic is running between host 2 and host 3.
   • In h1's xterm display, execute:

     ./send.py 10.0.3.3 6700 300
     

     This ensures that approx. 10 Mbps burst traffic is sent from host 1 to host host 3.
   • A few packets that causes microburst will be marked and received at the monitoring server (h4).
   • Similarly, burst traffic can be sent concurrently to multiple egress ports and the BurstRadar system will give desired results.

Result Screenshots