Investigating the efficiency of software-based switch implementations in physical networks

OData support
Dr. Csikor Levente
Department of Telecommunications and Media Informatics

The Software Defined Networking (SDN) is a constantly evolving, dynamic, manageable, easily adaptable and cost-effective network architecture that is ideal for today's bandwidth-consuming applications of many kind. The SDN concept separates the network control and the data transfer functions thus enabling the routing logic to be programmable and hiding the base network infrastructure from applications and services at the same time.

Unfortunately, there are only a few devices nowadays that support this concept even on the hardware side. Consequently, only software-based switch implementations are able to comply with the standardized total operation. One of the greatest advantages of the software-based implementations is that these can also be run on general-purpose hardware. However, the performance of the various implementations by real network requirements and traffic is nothing but obvious, especially in the most important contexts such as throughput, latency, energy consumption.

The goal of my thesis is firstly to develop performance evaluation methods for the various software-based implementations and secondly to analyse both the implementations and their effectiveness as result of real, physical network testing prepared along the aforementioned/developed evaluation methods.

In this thesis, I study the available free and/or open source software-based implementations of switch types and operations, and the tools to be used for generating real network traffic. I develop a uniform, comparable measurement methodology for the swift implementations and for the software. By means of the developed methodology, I conduct measurement on the performance of the individual implementations and I also analyse the optimization potential of the possible combinations. Finally, I provide my conclusion regarding the performance of the software based switch implementations.


Please sign in to download the files of this thesis.