A Study Of The Configuration Consistency In SDN

As a future network architecture, Software-Defined Networking(SDN) decouples the forwarding panel and the control panel, thus mitigates the problem of anomalies between network changes. The controller collects the topology and states of the network,determines the global routing policies and disseminates them to the switches. The switches forward packets according to the flow table. Although SDN centralizes the intelligence of the network into the controller, the packets are forwarded according to the flow table in each switch. Then the configuration inconsistency during updates is one main source of network instability. Even if the validity of the initial and final configurations is guaranteed, the interim complex and inconstant network states might cause routing conflicts and transmission disruption. Therefore, an e?cient updating scheme of configuration consistency is required.The problem of network configuration update has been studied for many years, but most of these solutions are used in the traditional network and protocols. Current solutions of SDN have several kinds: the tagging scheme, the order scheme and the dependency graphic scheme. The tagging scheme uses the tag to distinguish the old and new configurations. The packets are also added a tag to match the configuration. The order scheme finds a proper order to install update operation. The graphic scheme uses the resource allocation graph to analyse the relations between update operations, path, links and so on. The update process is slow and current schemes are not adaptable for incremental update. Flow-level consistent update in current schemes uses two-phase update and dividing rules periodically to guarantee the consistency, but it takes much time.In this paper, we provide a packet-level update scheme based on relation graph:analyse the relation between update operations, generate the relation graph and find a proper order of these operations. We propose a flow-level scheme based on K-prefix covering algorithm: the controller collects the existing flow information, computes K optimal prefixes covering the existing flows, then installs the new rule and K old sub-rules with lower and higher priorities respectively. We design the K-prefix covering algorithm by dynamic programming, which computes the K optimal prefixes with the minimum covering space. The experiment shows our scheme works well and accelerates the update process.