Modelo de tráfego em um cluster SDN ONOS/Atomix

Abstract

The software-defined network (SDN) is an emerging network paradigm that overcomes the limitations of a traditional network infrastructure. SDN can improve the monitoring, management, security and engineering of network traffic through a central controller. The SDN architecture with a single controller has disadvantages, such as low reliability and scalability. To solve these problems, the distributed SDN architecture emerged, where the network is controlled by a cluster with several controllers. To ensure the consistency of the network information, a consensus protocol is required to update the information and the synchronization of the controllers. There are two consistency models for SDN networks: eventual consistency and strong consistency. Strong consistency means that data readings on different controllers always produce the same result, while eventual consistency means that there are transient periods when readings on different controllers can produce different values. An example of an eventual consistency protocol is Anti-Entropy and a strong consistency is Raft. The consistency protocol chosen by each controller influences the amount of traffic between the controllers, which is used in the design of dimensioning the links between the controllers. In this work, an analytical model is developed in this work to determine the traffic between controllers as a function of the number of switches and links for a cluster of 3 controllers. Traffic analysis between ONOS controllers is made in two architectures: one with ONOS controllers, which will be called ONOS architecture and the other with ONOS controllers with the Atomix framework, which will be called ONOS/Atomix architecture. The analysis is done in two situations: with a different number of switches on each controller and with a different number of controllers. And the influence of traffic on the control plane on data plan traffic is also analyzed. The results show that the ONOS and ONOS/Atomix architecture have approximately the same traffic between controllers in different scenarios. Traffic grows directly proportional to the increase in the number of switches on the network and to the increase in the number of controllers.