Investigation of control plane in software-defined networks based on distributed system of virtualization functions

Автори: 
B.V. Koval, М.О. Seliuchenko, H.V. Melnyk, А.V. Kovalchuk

Lviv Polytechnic National University

In this paper are proposed the number of solutions for improvement of control plane in software-defined networks. One of the main problems that arise when designing and implementing new network is the problem of developing efficient and optimal network control system. SDN technology has control plane and data forwarding plane separated from each other. Decreasing of time of network adaptation is a major index of efficiency of software controlling system as it impacts on flexibility that is a main characteristic and advantage of SDN.  Improvement of control plane lies in creation of distributed system of virtualization functions (DSVF) which allows resolving major problems that arise when applying principles of centralized software control in carrier transport networks and increase control flexibility and network adaptability to unexpected events.
In order to manage network controller must have up-to-date information about all devices and their state and configuration in the network. Intensive exchange by monitoring messages impacts on the amount of signaling information that is transferred on the network. Moreover existing method of monitoring may cause delays in reaction of the control plane to extraordinary events that require immediate handling. Based on DSVF the method of dynamic monitoring for efficient management of network resources and decrease in signaling information amounts has been developed. It allows dynamically change monitoring frequency according to level of utilization of specific segment of the network. One of the main tasks of centralized architecture of SDN is ensuring of reliability and availability of controller. The function of decentralized management that forms reserved control plane and ensures network reliability has been proposed. When programming flow tables sequentially network reconfiguration duration can significantly vary depending on the number of switches to be reconfigured, number of flow tables on each switch and number of flows to be modified in each flow table. Function of parallel programming of switch flow tables for increasing of network reconfiguration speed has been proposed. Key novelty of DSVF lies in the fact that management processes on each independent switch are partly performed by applications that are installed on each switch. Each application executes one specific function from the set of functions that have been created for a specific case in the network and in general this applications form distributed system of virtualization functions. All solutions have been tested based on the developed imitation model. The results of its work confirm efficiency and practical value when applying proposed solutions on a real transport network.