Most of today's computer networks are highly decentralized. Each network element operates completely autonomously using its own control software. Network elements communicate with each other and detect each other's state using a variety of complex protocols that enable the network to operate properly. The only centralized point of such networks is the management plane, although it is not essential either, since autonomous management is provided by all network elements.
In recent years we have witnessed a rapid emergence of a totally different network concept. The network of the future is highly centralized. The centralized controller's software of such network defines the behaviour of every network element at any time. In this way the whole network is defined by the software, through a single control point. The concept is known as »Software Defined Networks« (SDN).
In the master thesis we describe the architecture and principle of operations of software defined networks in detail. We substantiate the advantages and disadvantages of such networks. The main part of the thesis is a proposal of a new, more optimal, method for controlling SDN. The method brings significant improvements of certain important weaknesses of software defined networks, such as poor performance and scalability of the controller and the lack of compliance with real-time requirements.
The proposed method comprises a universal definition of the control message, which enables the establishment of a new flow forwarding path through multiple network elements. The modification of the controller makes it possible to cast such a message so that it reaches all relevant elements or SDN switches. Moreover, adjustable design of the switches further contributes to proper interpretation of these messages.
Additionally, we present a hypothetical network which was used as a basis for theoretical calculations of efficiency improvement of the new method, compared to the existing ones. The calculations show a substantial improvement of the aforementioned drawbacks of software defined networks.
Finally, we have put up a simulation environment for testing of the new method. The performed simulations have showed improvement in most network characteristics. Deviations between the simulated and real network environments have come into consideration with the other characteristics. Consequently, by running the simulation it was not possible to demonstrate all the advantages of the proposed control method.
|
