Congestion control in a software defined network

Abstract

Cisco predicts that by 2021, more than 80% of all network traffic will be video. With the introduction of applications that allow for easier consumption of media content, networks must be able to handle the load generated from this traffic. Conventional methods rely on hardware efficiency, which becomes harder and harder as physical or environmental limitations are reached (such as the number of transistors that can fit on a single chip). Efficient congestion control is a key component for enabling IP multicast deployment over the Internet. Software Defined Networks (SDNs) allow for a separation of the control and data planes, and a centralized control of the networks operations through software. SDNs also enable the use of non-proprietary hardware, allowing for more flexibility to network engineers. Network devices communicate with the controller, which carries out all the networking decisions. This project proposes a testbed for a small SDN to control multicast video in different scenarios, consisting of a controller, a switch and 3 hosts - one acting as producer of the multicast stream while the 2 others as consumers. The controller and the switch communicate through the OpenFlow protocol, allowing for decisions made by the controller to be pushed onto the switch. By modifying the OpenFlow packet headers, it is shown how flow control can be achieved automatically as well as quickly when faced with conditions catered for by the programmer. This project also confirms the importance of simulation software, allowing for fast and accurate representations of real hardware while avoiding the cost and maintenance that comes with said hardware. Additionally, commands executed on simulated hardware as well as the code used can be executed on real hardware and the effect would be the same. Finally, physical access points are connected to the testbed and it was demonstrated that flows by the controller can also affect clients connected to these access points by replicating the various scenarios addressed with virtualized clients to wireless clients, highlighting practical applications of this project.