Software-Defined Networking (SDN) provides a more sophisticated and flexible architecture for managing and monitoring network traffic. SDN moves part of the decision-making logic (i.e., flow processing and packet routing) from network devices into a logically centralized controller. However, the expected behavior and configuration of network devices are often defined directly in the controller as static rules for specific situations. This approach becomes an issue when associated with an increasing number of network elements, links, and services, resulting in a large amount of rules and a high overhead related to network configuration. As an alternative, techniques such as Policy-Based Network Management (PBNM) and more specifically policy refinement can be used by operators to write Service Level Agreements (SLAs) in a user-friendly interface without the need to manually reconfigure each network device. To address these issues, we specifically introduce ARKHAM: an Advanced Refinement Toolkit for Handling SLAs in SDN. In this article, we present (i) a policy authoring framework that uses logical reasoning for the specification of business-level goals and to automate their refinement; (ii) an OpenFlow controller which performs information gathering and configuration deployment; (iii) a policy repository that stores information about the behavior of the infrastructure, which is obtained by the OpenFlow Controller, and policy authoring operations; and (iv) a formal representation using event calculus that describes our solution. The main contributions of this work are (i) the capacity to deploy refined policies with minimal human intervention; (ii) analysis of the infrastructure's ability to fulfill the requirements of high-level policies; (iii) decreased amount of network rules coded into the controller; and (iv) management and deployment of new rules with minimal disruption to the network. The experimental results demonstrate that the refinement toolkit achieves the expected results within acceptable performance bounds, even with the increasing complexity and size of SLAs, network topologies, and repositories.
Read full abstract