High Availability and Scalability Schemes for Software- Defined Networks (SDN)

Abstract

A proliferation of network-enabled devices and network-intensive applications require the underlying networks not only to be agile despite of complex and heterogeneous environments, but also to be highly available and scalable in order to guarantee service integrity and continuity. The Software-Defined Network (SDN) has recently emerged to address the problem of the ossified Internet protocol architecture and to enable agile and flexible network evolvement. SDN, however, heavily relies on control messages between a controller and the forwarding devices for the network operation. Thus, it becomes even more critical to guarantee network high availability (HA) and scalability between a controller and its forwarding devices in the SDN architecture. In this dissertation, we address HA and scalability issues that are inherent in the current OpenFlow specification and SDN architecture; and solve the problems using practical techniques. With extensive experiments using real systems, we have identified that iii the significant issues of HA and scalability in operations of a SDN such as single point of failure of multiple logical connections, multiple redundant configuration, unrecoverable interconnection failure, interface flapping, new flow attack, and event storm. We have designed and implemented the management frameworks that deal with SDN HA and scalability issues that we have observed from a real system. The proposed frameworks include various SDN HA and scalability strategies. For SDN HA, we have developed several SDN control path HA algorithms such as ensuring logical control path redundancy, transparency of a controller cluster, and fast and accurate failure detection. We validate the functionalities of the proposed SDN HA schemes with real network experiments. The proposed SDN control path HA algorithms overcome the limitations of the current Open- Flow specification and enhance performance as well as simplify management of SDN control path HA. For SDN scalability, we have proposed and developed our management framework in two different platforms; an embedded approach in the OpenFlow switch and an agent-based approach with the SUMA platform that is located near the Open- Flow switch. These platforms include various algorithms that enhance scalability of SDN such as Detect and Mitigate Abnormality (DMA), Modify and Annotate Control (MAC), and Message Prioritization and Classification (MPC). We have shown that the proposed framework effectively detects and filters malicious and abnormal network behaviors such as new flow attack, interface flapping, and event storm.